XEE100 All Slides

IoT Intro
Olav Solgaard
Bio
Olav Solgaard earned his Ph.D. degree from Stanford University in 1992. His doctoral dissertation:
“Integrated Semiconductor Light Modulators for Fiber-optic and Display Applications” was the basis
for the establishment of a Silicon Valley firm Silicon Light Machines (SLM), co-founded by Dr.
Solgaard in 1994. From 1992 to 1995 he carried out research on optical MEMS as a Postdoctoral
Fellow at the University of California, Berkeley, and in 1995, he joined the Electrical Engineering
faculty of the University of California, Davis. His work at UC Davis led to the invention of the multi-
wavelength, fiber-optical switch, which has been developed into commercial products by several
companies. In 1999 he joined Stanford University where he is now a Professor of Electrical
Engineering and the Director of Graduate Studies in the Department of Electrical Engineering.
Professor Solgaard’s research interests include optical MEMS, Photonic Crystals, optical sensors, microendoscopy, atomic force
microscopy, and solar energy conversion. He has authored more than 350 technical publications and holds 70 patents. Professor
Solgaard is a Fellow of the IEEE, the Optical Society of America, the Royal Norwegian Society of Sciences and Letters, and the
Norwegian Academy of Technological Sciences.
Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

IoT Overview
• What are the disciplines of IoT?
• Cool applications and fantastic opportunities – Ada Poon
• Sensors – Beth Pruitt
• Small, inexpensive and very, very good
• Circuits – Boris Murmann
• This is where it started
• Embedded systems – Phil Levis
• Sensors, circuits and communication links playing together
• Networks – Ayfer Özgür-Aydin
• How does the internet support IoT?
• Big data and machine learning
• Other certificates focus on these aspects of IoT
Ideal IoT
• Ubiquitous
Why is it important?
• Smart
• Fourth industrial • Agile
revolution • On demand
• Blend into background
• IoT will touch everything
• Secure
• Everything will have IoT
• Low maintenance
• Fast
• Upgradable
• Growing
• Adaptable

IoT will be the “next big thing”
Tsai, 2014 International Solid-State Circuits Conference (ISSCC), Keynote Talk
What is IoT? Cool

Locomotive for
innovation and
applications
economic growth
Big data
Sensors
Internet
Security
Circuits
Sensors
Sensors

IoT Overview
• What will you learn in the short course?
• These are the questions we will help you answer:
• What is your perspective on IoT?
• How does the disciplines of IoT overlap with your expertise and interests?
• Where are the holes in your knowledge that you want to fill?
• What IoT opportunities do you want to pursue?
• And ultimately
• How to get the most out of your IoT certificate
Cool Applications – Professor Ada Poon

• Sustain (Smart cities)
• Move (Self driving cars)
• Heal (Healthcare)
• Feed (Agriculture)
• Make (Manufacturing and packaging)
• The purpose of this section is two fold:

• Get you excited about IoT potential
• And more importantly, have you thinking of better ideas than the ones
already out there

Smart Waste Management
• Go from collecting 840

containers 4 times a day to
collecting 80 containers a
day.
• Increase waste collection
efficiency by 90%.
Dublin Airport
http://ecubelabs.com/case-studies/dublin-airport/
Smart Street Lights
• Save energy
• Reduce light pollution
• Faster replacement
http://www.tvilight.com/

Smart Street Parking
• Infrared- and magnetic-based vehicle detection
sensor mounted on the road surface
• Zigbee, LoRaWAN wireless connections
• Mesh networks are implemented with
in street lights.
• Apps to direct drivers to
empty spaces
• Dynamic parking pricess
http://www.libelium.com/smart_parking/
Connected Vehicles – Combine private and collective

transportation: The best of both
http://government-2020.dupress.com/trend/connected-vehicles/

Mood Enhancing – Sleep Monitoring
https://www.beddit.com/
Enhance Adherence – Ingestible Sensors
http://www.proteus.com/
http://www.sfchronicle.com/business/article/Did-you-take-your-pill-Ingestible-sensors-can-11206980.php

Challenges
• A lack of Electronic Health Record (EHR) integration and concerns about data
security may prevent healthcare from fully adopting the IoT technology.
• The need to adopt an integration-first mindset instead of keep building
interesting/fun gadgets. Sometimes, a dumb gadget can be as useful if it could
integrate seamlessly with the EHR.
• False positives
• Doctor’s offices will be filled with perfectly healthy, but anxious patients
Precision Agriculture
https://www.accenture.com/cn-en/insight-accenture-digital-agriculture-solutions

Food Safety & Storage Efficiency
• Wifi or cellular connectivity

• When a produce recall is
initiated, the juice machine
will check the packs and
prevent the machine from
pressing affected packs.
https://www.juicero.com/
Smart Manufacturing
https://www.slideshare.net/andrejt/ntk-2015-internet-of-things-track-iot-smart-home

Scaling of sensors
• Many sensors can be modeled as simple
Reference
mechanical harmonic oscillators frame
• The figure shows an accelerometer
• The proof mass (m) is accelerated and displaced k b
• The displacement (Dz) of m is measured
• The acceleration is calculated as a=k·Dz/m (The
spring constant, k, is known) m
• The highest frequency that this accelerometer z
# &
can measure is 𝑓 = Accelerometer modeled as
$% '
mechanical harmonic oscillator
Capacitive measurement of displacement

• Most IoT sensors measure changes in
electrical properties:
• Resistance, capacitance, inductance
• Our accelerometer is completed by I k
capacitive measurement of displacement m
+ +
• Capacitors are great at measuring short Vin g z
V E
distance with very good absolute -
-
accuracy
• Measurements of large displacements are
better done by inductors (magnetic fields)

Block diagram of MEMS accelerometer
Noise and disturbances enter the system both as an equivalent acceleration noise,
and as an equivalent signal noise in the position sensor.
Disturbances
and Noise
+
Proof
mass Position
Acceleration
+
+
Position Noise
Sensor +
MEMS sensors: Small is beautiful

k 1 k
Dz = (2p × f ) × Dz
2
a= f =
m 2p m
• Design objective: Measure the smallest possible a up to a given f
• Looking at the first part of the equation, we might think that we should use a low k and large m
– => low f
– Given a constant f, only the ratio k/m~f2 is important
– A miniaturized accelerometer must have a small proof mass and a weak spring, but it is the
ratio of the two that is important!
– To measure small accelerations up to a given f, we therefore have to measure Dz very
accurately
– For application specific minimum detectable acceleration and bandwidth, there is a
minimum detectable displacement that our position sensor must be able to measure
• Small structures have position sensitivity and can be designed to have any resonance frequency
• => MEMS sensors!

Electrical components used in miniaturized sensors
Generic block diagram
Wireless Accelerators &

transceiver Microcontroller
Mixed-signal Sensors
interfaces Actuators
Energy
Memory
Management
1000 mAh
11.4 µA

Key: Duty-cycling
§ IoT transceivers are “mostly” off

§ Example: Receive/transmit within 1 ms, then go to sleep for 1 sec
http://www.mouser.com/applications/low-power-ewc-design/
Power/energy cost of wireless transmission

Bluetooth
Nike+ ZigBee WIFI
LE Ant-sized radio
Power
0.147 0.675 35.7 210
(mW)
Bits/sec 960 272 192 40M
Energy/bit
153 2480 186,000 5.25
(nJ)
§ Numbers are a strong function of reach
§ Energy per bit tends to be lower for high-data rate links à Invested power
amortizes better
https://www.digikey.com/en/articles/techzone/2011/aug/comparing-low-power-wireless-technologies

Key issue for sensors: 1/f noise
offset
+ drift
thermal noise
signal
fcorner log f
§ Analog MOS circuits come with a variety of low frequency artifacts that can
easily ʺswamp outʺ DC or low-frequency signals of interest
What is an embedded system?
“An embedded system is a computerized system that

is purpose-built for its application.”
Elicia White
Making Embedded Systems, O’Reilly
This has deep implications for how they are designed and why.
Typically, want to minimize cost/maximize lifetime for a given expected workload.
Optimized, custom software that uses as few resources as possible.

Two basic cost considerations dominate
Energy
§ Embedded systems have an expected workload – want to minimize energy
required to handle the workload (longer battery life, more efficient)
§ More powerful, highly featured MCUs draw more power: can be more power
efficient, but consume more energy for a given workload
Money
§ Many embedded systems markets have tight margins (cars, appliances, etc.)
§ More powerful, highly featured MCUs cost more
Both costs push designs to pick minimal MCU and optimize code for it
IoT: MGC Architecture

eMbedded Gateways Cloud
devices
6lowpan, 3G/4G
ZigBee, Ethernet
ZWave,
Bluetooth, TCP/IP
WiFi,
WirelessHART

Two Game-Changers
ARM Cortex M series

§ First released 2004
§ Ultra-low power 32-bit processor
§ 8-96kB of RAM, 64-512kB code flash
§ Sleep currents recently dropped <1µA
Bluetooth Low Energy

§ First released in 2006
§ Send a 30 byte packet once per second, last for a year
on a coin cell battery
§ Support was weak until Apple incorporated into
iBeacon, now all major smartphones include it

What is IoT? Cool
Locomotive for
innovation and
applications
economic growth
Big data
Sensors
Internet
Security
Circuits
Sensors
Sensors

IoT Overview
• What will you learn in the short course?
• These are the questions we will help you answer:
• What is your perspective on IoT?
• How does the disciplines of IoT overlap with your expertise and interests?
• Where are the holes in your knowledge that you want to fill?
• What IoT opportunities do you want to pursue?
• And ultimately
• How to get the most out of your IoT certificate

Cool Applications
Professor Ada Poon
Outline
• Smart cities
• Healthcare
• Agriculture
• Manufacturing and logistics

Source: Verizon. Smart Cities Solutions, 2014
Dublin Airport New York City

http://ecubelabs.com/ http://bigbelly.com/

• Ultrasonic fill-level sensor
• Cellular IoT
• Solar powered
• Predictive pattern recognition
http://ecubelabs.com/integrated-waste-management/
• Go from collecting 840

containers 4 times a day to
collecting 80 containers a
day.
• Increase waste collection
efficiency by 90%.
Dublin Airport
http://ecubelabs.com/case-studies/dublin-airport/

Smart Street Lights
• Save energy
• Reduce light pollution
• Faster replacement
http://www.tvilight.com/
Smart Street Lights

• Light sensor, motion sensor
• Cellular IoT
• Real-time mesh network
https://chess-wise.eu/en/smart-street-lighting/

• In the past year, how many times did you give up when looking for a parking
space in cities? How many times did you argue with someone about a parking
spot? On average, how long did you take to look for a parking spot?
• More than 30% of a city’s traffic is caused by drivers searching for a parking
spot.
• In New York City, 29% of commuters said that they spent 20 minutes on
average looking for a parking spot and 10% spent more than 40 minutes.
http://www.govtech.com/transportation/Smart-Parking-Tech-US-Cities.html

• Infrared- and magnetic-based vehicle detection
sensor mounted on the road surface
• Zigbee, LoRaWAN wireless connections
• Mesh networks are implemented with
in street lights.
• Apps to direct drivers to
empty spaces
• Dynamic parking prices
http://www.libelium.com/smart_parking/

Smart Street Parking in Poland
http://www.worldsensing.com/success-story/gliwice-smart-parking-solution/#
Security without Surveillance
http://www.sensity.com/for-security-1/

Security without Surveillance
• Real-time analytics rather than human-monitored surveillance
• Edge-based analytics rather than cloud-based analytics
• All videos are stored locally.
• This also reduces the requirement on datarate.
• Resultant analytics sent to central cloud database for issuing alerts
Connected Vehicles
http://government-2020.dupress.com/trend/connected-vehicles/

Connected Vehicles – Stages of Safety Innovation
• Passive vehicle safety uses sensors to
take the vehicle’s immediate
surroundings into consideration.
• Recent efforts enable the sharing of
information gathered by the sensors
between vehicles, and between vehicles
and their surroundings to increase safety
further.
• V2X (vehicle-to-X, where X
represents other vehicles,
infrastructure, roads, and so on)
• A step towards autonomous driving
https://dupress.deloitte.com/dup-us-en/deloitte-review/issue-12/the-rise-of-safety-innovations-in-intelligent-mobility.html
Let’s Recap
Sensors
• Ultrasonic fill-level sensor
• Light and motion sensors
• Infrared- and magnetic-based vehicle detection sensor
• Camera
Communication Protocols
• Cellular IoT (LTE Cat-M1 and NB-IOT)
• Zigbee, LoRaWAN
• V2X
Networking Protocols
• Mesh network

Outline
• Baby monitoring
• Elderly monitoring
• Mood enhancement
• Disease treatment and progression monitoring
• Enhance adherence
Baby Monitoring – Activity Tracking

• Body position
• Breathing
• Oxygen level
• Skin temp
• Wake/sleep pattern
• ECG
Can be integrated with other IoT devices such as thermostat and camera to “close
the loop”. For example, if the baby is too warm, the thermostat will
automatically adjust.

http://mimobaby.com/ http://www.owletcare.com/smart-sock-2/
1. The turtle sends information about

the baby’s breathing, body position,
sleep activity, and skin temp to the
Lilypad via Bluetooth LE.
2. The Lilypad streams data and live
audio to the cloud via WiFi.
3. Parents receive real-time insight
about their baby on their
smartphone.
http://mimobaby.com/

Baby Monitoring – Urinary Tract Infection Monitoring
• Urinary tract infection (UTI) is
the second most common
infection in the US accounting for
7M hospital visits and 100,000
hospitalization per year.
• It is easy to cure if detected and
treated in early stage.
• Urine culture test is accurate but
time consuming. Dip stick test is
fast but high false alarm rate.
https://sites.google.com/site/jagpurdue/projects/catalyst
Baby Monitoring – Urinary Tract Infection Monitoring
• Urine-activated paper battery (self-powered)

• Paper-based colorimetric nitrite sensor consisting of an LED, a urine-
absorbing strip, a reagent strip, an active photodiode, and a reference
photodiode.
• Sensor signal is converted into a PWM waveform.
• BLE module transmits the PWM signal to the caregiver.
https://sites.google.com/site/jagpurdue/projects/catalyst

Elderly Monitoring – Incontinence Management
• In most nursing homes, between 40% and

60% of residents suffer from urinary
incontinence.
• Smart diaper allows caregivers to
remotely detect if an incontinence event
has occurred.
• Improved quality and dignity of care by
not having to disturb the elderly.
http://rfmicron.com/health-care/ http://www.medisens.com/news/press-release-1
Elderly Monitoring – Fall Detection

• One-fourth of Americans aged 65+ falls each year.
• Every 11 seconds, an older adult is treated in the emergency room for a fall;
every 19 minutes, an older adult dies from a fall.
• Falls are the leading cause of fatal injury and the most common cause of
nonfatal trauma-related hospital admissions among older adults.
• Falls result in more than 2.8 million injuries treated in emergency departments
annually, including over 800,000 hospitalizations and more than 27,000
deaths.
• In 2013, the total cost of fall injuries was $34 billion.
• The financial toll for older adult falls is expected to increase as the population
ages and may reach $67.7 billion by 2020. (The medicare budget is $584
billion in 2016.)
https://www.ncoa.org/news/resources-for-reporters/get-the-facts/falls-prevention-facts/

Elderly Monitoring – Fall Detection
http://www.semtech.com/wireless-rf/internet-of-things/downloads/Semtech_Health_FallDetection_AppBrief-FINAL.pdf
Mood Enhancement
• Relaxing music could be cued to ease stress.
• Window shades could be programmed to let in the maximum amount of
natural light.
• Use IoT to encourage healthy behaviors.
• Automatically dim the lights in the home at a recommended bedtime.
• Automatically turn off the TV to encourage exercise.

Mood Enhancing – Sleep Monitoring
https://www.beddit.com/
Disease Treatment and Progression Monitoring – Parkinson’s
Patients with Parkinson’s disease must be continually assessed in order to keep

up with their symptoms. This becomes potentially problematic as symptoms
fluctuate on a constant basis, and a monthly check in with their doctor may not be
representative of their experience.
https://www.slideshare.net/rcnossen/healthcare-iot-and-analytics-to-treat-parkinsons

Instead of patients actively performing certain tasks, could we monitor disease

progression passively in the background?
http://parkinsonmpower.org/

• IoT opens up new possibilities for disease treatment through remote
monitoring. Using sensors, mobile devices, and advanced machine learning
capabilities, a patient could keep track of a host of valuable data from mobility
to sleep patterns all in real time. This information will give practitioners a
more complete look into the progression of their patient’s disease states.
• Intel and Michael J. Fox Foundation for Parkinson’s Research, and Pfizer and
IBM are individually collaborating on this idea. The collaboration involves
planned clinical trial.

Enhance Adherence
• 84% of U.S. healthcare spending is on patients with chronic conditions.
• More than 50% of prescribed medications are not taken as directed.
• Reasons why people are not able to take their medication as directed:
• They may forget.
• They may not be convinced of the medication’s effectiveness or be unsure
that it is working.
• They may fear the side effects or have difficulty taking the medication.
• The rising cost of prescription medications is a barrier for many.
Anderson G. Chronic conditions: making the case for ongoing care. Baltimore, MD: Johns Hopkins University, 2010.
Sabate E. Adherence to long-term therapies: evidence for action. World Health Organization, 2003.
Enhance Adherence – Ingestible Sensors
http://www.proteus.com/
http://www.sfchronicle.com/business/article/Did-you-take-your-pill-Ingestible-sensors-can-11206980.php

Challenges
• A lack of Electronic Health Record (EHR) integration and concerns about data
security may prevent healthcare from fully adopting the IoT technology.
• The need to adopt an integration-first mindset instead of keep building
interesting/fun gadgets. Sometimes, a dump gadget can be as useful if it could
integrate seamlessly with the EHR.
Outline
• Precision agriculture
• Connected livestock
• Food safety

• A farming management concept based on observing, measuring and
responding to inter and intra-field variability in crops.
• In the past, precision agriculture technology was implemented by big
agribusinesses due to high costs.
• IoT technologies – which include everything from GPS services, sensors, and
big data calculation – have made precision agriculture affordable by many
farmers.
• Farmers don’t have to rely as much on their gut. Instead, they can make
decisions based on detailed information about water, climate changes, soil
quality, the health of their crops and livestock, and the conditions of their
machinery.
https://flex.com/intelligence/iot/old-macdonalds-new-connected-farm
https://www.accenture.com/cn-en/insight-accenture-digital-agriculture-solutions

Opportunities for
vertical integration
https://www.cbinsights.com/research/agriculture-tech-market-map-company-list/
Connected Livestock
• Around 1.4 billion cattle around the world
• Animals can't tell you when they first get sick. It can be hard for humans to
tell a cow is ill until there are visible signs of sickness.
• IoT sensors cannot diagnose an illness but it will let the farmer know when
something needs attention.
https://www.smaxtec.com/en/

Connected Livestock
• pH measurement
• Early detection of fermentation disorders
• Activity level measurement
• Early, automatic oestrus detection
• Onset illness
• Temperature measurement
• Early detection of onset of diseases such as feverish disorders, metabolic
disorders, post-calving disorders
• Early detection of start of calving
https://www.smaxtec.com/en/
Connected Livestock
http://uk.smaxtec.com/smaxtec-system/

Food Safety
• Wifi or cellular connectivity

• When a produce recall is
initiated, the juice machine
will check the packs and
prevent the machine from
pressing affected packs.
https://www.juicero.com/
Outline
• Smart manufacturing
• Smart packaging

Smart Manufacturing
• The use of IoT devices to improve efficiency and productivity of
manufacturing operations. Typically, it involves retrofitting sensors to
existing manufacturing equipment. But new manufacturing equipment often
comes with IoT sensors pre-installed.
• According to IDC data, published early 2017, the manufacturing industry was
good for a total IoT spending of $178 billion in 2016, which is more than
twice as much the second largest vertical market, transportation.
• Manufacturing operations accounts for 57.5% of the total IoT spending on
manufacturing.
http://www.businesswire.com/news/home/20170104005270/en/Internet-Spending-Forecast-Grow-17.9-2016-Led
Smart Packaging
• Packaging systems used with food and pharmaceutical that help extend shelf
life, monitor freshness, display information on quality, improve safety, and
improve convenience.
• Usually involve active functions beyond the inert passive containment, for
example, the ability to sense or measure an attribute of the product, the inner
atmosphere of the package, or the shipping environment. This information can
be communicated to users or can trigger other active packaging functions.

Conclusion
http://www.eetimes.com/author.asp?doc_id=1328602

Introduction to Sensors
Professor Beth Pruitt
Stanford University
This modules provides an Introduction to Sensors and includes an introduction to

sensors as transducers from physical parameters to signals; a review of sensor
terminology; and discusses sensor selection including how to read a specification
sheet. If you want more detail, check out the course ENGR/ME220 which covers
enough about all kinds of sensors so that you can make good decisions about
recommending, selecting and using sensors in projects, products and research.
Learning Objectives
• Describe and define performance criteria for sensors (e.g., linearity, sensitivity,
resolution, noise)
• Explain the operating mode for some common IoT transducers and sensors
(strain gage, accelerometer, gyros, temperature, pressure sensors…)
• Interpret a specification sheet and extrapolate missing performance data

What is a sensor?
input -> Sensor -> output

Sensation
Imagine Life Without…
Touch
Hearing
Propioception
Vision
Gun Legler
Perception
Force or
of Topography or
Displacement
Stiffness
Rodin’s Cathedral

external load
skin mechanics
stress/strain at neuron
neuron mechanics,
channel id & biophysics,
cell physiology
electrochemical response
neural circuit
past/current stimuli
behavioral response
SKIN: your largest and fastest growing organ
Barrier Layer:
• Respiration/perspiration
• Cooling
skin surface
Sensory Function:
• Static Forces
• Dynamic Forces 10’s-100’s µm
• Temperature
100’s-1000’s µm

Digital Signal Encoding : Amplitude coded by Pulse Density
Adaptation : Pulse density decays over time

Quantifiying touch sensation
• Dynamic Range – what range of force or deformation can you feel?
• Bandwidth – what frequency or on/off rate can you detect?
• Sensitivity – how strong is the output signal relative to the input? Is it the
relationship linear?
• Resolution – what is the smallest feature or smallest force you can detect?
Sensor Terminology

Transducer, n. Any device by which variations in one physical quantity (e.g.
pressure, brightness) are quantitatively converted into variations in another (e.g.
voltage, position).
OR…a device for which changes in input quantity A produce corresponding, predictable
changes in output quantity B
For our purposes, Sensors convert physical to electrical signals
Generic Sensing Application
Input
(measurand)

The Ideal Sensor
Sensor
Output
Measurand
Sensor Specifications

Sensitivity
Sensor
Output
Measurand
Sensor
Output
Measurand
Non-Linearity
Voltage
Pressure

Non-Linearity
slope1
slope A
Voltage slope B
Pressure
Hysteresis
Voltage
Pressure

Repeatability
Voltage
Pressure
Drift
Same pressure, take

measurement for a long time
Voltage
Pressure

Specifying Non-Linearity
% full
scale
output
Voltage
Pressure
Span / Range
linear saturation
Voltage
Pressure

Sensor Dynamics
2nd order Mechanical system

• Mass: F=ma b
F(t)
m
• Damper: F=bv k
x(t)
x(0)=0
• Spring: F=kx

Transfer function & characteristic eqn
d2x dx
f (t) = m 2 + b + kx
dt dt
X(s) 1/m
= G(s) = 2
F(s) s + bms+ k m
X(s) Kω n 2
= G(s) = 2
F(s) s + 2ζω n s + ω n 2
Step Response
Underdamped
Critically damped
Final
value
Voltage
Overdamped
Time

Sensor stimulus
Dynamics
Ideal
Slight delay
decay
combo
2nd order
system
time
Bandwidth
Δt1
Δt2
delay
decay
time
delay
decay
Bandwidth
1/Δt2 1/Δt1
frequency

Linearization and Error
All models are wrong, some are useful.

- George Box
Overall Error (Error Band)
Error
Correct Value
Band
at Output
Sensor
Output
Measurand
True Value of
Measurand

x
x
Accuracy vs. Precision

x x
x
Imprecise and Inaccurate
• Accurate x xx
x
x
• Average of sampled output is close to
real value (AC errors and noise)
• Precise Precise but Inaccurate
• Sampled output is consistently tightly
grouped with consistent offset from real x
value (DC errors) x

x x
x
Accurate but Imprecise
x x
x
x x
Precise and Accurate
Linearity of a resistive sensor in a half bridge

Vin
linear
R
Vout
Vout
Rsense
Rsense
æ Rsense ö
Vout = Vinç ÷
è Rsense + R ø

Linearity details
æ Rsense ö
Vout = Vinç ÷ = V (Rs )
è Rsense + R ø
We’re usually interested in the behavior for relatively small changes in Rs.
Rs = Ro + DR
Linearity analysis using a Taylor expansion

æ Rsense ö
è Rsense + R ø
We’re usually interested in the behavior for relatively small changes in
Rs so evaluate the linear (sensitivity) and non-linear (error) terms at an
operating point – relative to starting value R0
Rs = Ro + DR
Recall that the Taylor series approximation of f(x) about operating point a is defined as:
offset Linear term Non-linear error terms

Linearity analysis using Taylor expansion
æ Rsense ö
è Rsense + R ø
Rs = Ro + DR
é dV ( Rs ) ù (DR) 2 é d 2V ( Rs ) ù (DR)3 é d 3V ( Rs ) ù
V ( R s ) = V ( Ro ) + DR ê ú + ê 2 ú + ê ú ...
ë dRs û Rs = Ro 2 ë dRs û Rs = Ro 6 ë dRs 3 û Rs = Ro
Ro Ro (DR) 2 Ro (DR)3 Ro
V ( R s ) = Vin + DR - +
R + Ro ( R + Ro ) 2
4 ( R + Ro ) 3
36 ( R + Ro ) 4
offset Linear term Non-linear error terms
“Sensitivity” Decreasing weight with increasing

exponent
Quadratic/linear =

Why do we care so much about Linearity?
linear
Vout
Rsense
1- (cos2wt) 2
Vibration rectification – offset errors that

add up and don’t average out to zero

Noise
Electronic Noise Sources

• EMF - capacitive & inductive pickup
• Johnson noise
• All resistors and dissipative systems
• Thermal/Brownian random molecular interactions VJ = 4k B TR
• 1/f noise (shot, flicker, Hooge)
• Semiconductor based electronics, amplifiers, instruments αVR2
• Seminconductor resistors, Hooge noise VH =
Nf
• Drift
• Accumulated offset errors
• Very low frequency fluctuations?

Other electronic noise
• Shot noise
• Associated with pn junctions Sn ( f ) = 2qe IDC
• Flicker noise (also 1/f)

KIDC
• Associated with trap charge states in diodes and FETs Sn ( f ) = 2qe IDC +
• Amplifier noise f
• Multiple factors, depends on op-amp type, see Senturia
• Example: AD624 instrumentation amplifiers have
• 4nV/rtHz voltage noise at low frequencies
• 200fA/rtHz current noise above 10Hz
Intrinsic thermal noise

mechanical harmonic oscillator
w0 1
x˙˙ + x˙ + w 0 x = [Fsignal (t) + Fnoise (t)]
2
k Q m
[m Hz]
k
m w0 = m 4kB Tk 2
Sn force ( f ) =
ω 0Q
4k B T
Cantilever
t w Sn position ( f ) =
kω 0Q
⎛ ω 2 ⎞2 ω2
[m Hz]
2
⎜1− 2 ⎟ + 2 2
L ⎝ ω0 ⎠ Q ω0
Quality factor, Q =(mk)1/2/b, for a cantilever ~30-50 in air

Resolution
• Smallest signal the transducer can
resolve or produce for a particular set
Voltage (V)
of operating conditions
• Noise has some frequency distribution Noise
Floor
• Signal to noise ratio (SNR)
• Depends on bandwidth
Time (s)
Noise _ Density
RESOLUTION =
Sensitivity
Integrated _ Voltage _ Noise
MIN =
Sensitivity
Scaling Issues for Force Sensors
• Noise generally does not improve with miniaturization

• Electrical Noise will be about same
• Energy Fluctuations (kBT) will result in larger errors
• Thermomechanical noise scales as 1/√m
• Some 1/f noise becomes more significant
• Increased surface/volume

Reading a Spec Sheet

0.1 mW to 1 mW power – also way too high for a phone

Basic Description
Key features:
Outputs fused sensor data Quaternion, Euler angles, Rotation vector,
Linear acceleration, Gravity, Heading
3 sensors in one device an advanced triaxial 16bit gyroscope, a versatile,
leading edge triaxial 14bit accelerometer and a
full performance geomagnetic sensor
Data sheet
Small package LGA package 28 pins
Footprint 3.8 x 5.2 mm², height 1.13 mm²
BNO055
Power Management Intelligent Power Management: normal,
low power and suspend mode available
Intelligent 9-axis absolute orientation sensor

Common voltage supplies
Digital interface
VDD voltage range: 2.4V to 3.6V
HID-I2C (Windows 8 compatible), I²C, UART
VDDIO voltage range: 1.7V to 3.6V
Consumer electronics suite MSL1, RoHS compliant, halogen-free
Bosch Sensortec Operating temperature: -40°C ... +85°C
Key features of integrated sensors:

BNO055
Page 13
Accelerometer features Data sheet
Programmable functionality Acceleration ranges ±2g/±4g/±8g/±16g

Low-pass
1.2 Electrical and physical characteristics, measurement filter bandwidths
performance 1kHz - <8Hz
Operation modes:
Table 0-2: Electrical characteristics BNO055 - Normal
- Suspend
OPERATING C ONDITIONS BNO055 - Low power
Parameter Symbol Condition Min Typ- Standby
Max Unit
St art -Up t ime T From Off t o configurat ion mode

Sup 400- Deep suspend ms
POR t ime
On-chipT interrupt controller
From Reset t o Normal mode
POR
Motion-triggered
650
interrupt-signal
ms
generation for
- any-motion (slope) detection
Dat a Rat e DR
- slow or no motion recognition
s. Par. Fusion Out put dat a rat es
Dat a rat e t olerance DR tol ±1- high-g detection

%
9DOF @100Hz out put
dat a rat e
(if int ernal oscillat or is
used)
OPERATING C ONDITIONS ACCELEROMETER

Parameter Symbol Condition Min Typ Max Units
Accelerat ion Range gFS2g Select able ±2 g

via serial digit al int erface
gFS4g ±4 g
gFS8g ±8 g
gFS16g ±16 g
OUTPUT SIGNAL ACCELEROMETER

(ACCELEROMETER ONLY MODE )
BST-BNO055-DS000-12 | Revi si on 1. 2 | November 2014 Bosch Sensortec
© Bosch Sensortec GmbH reserves al l ri ghts even i n the event of i ndustri al property ri ghts. We reserve al l ri ghts of di sposal su ch as copyi ng and Digital from
Sensit
passiivitngyon to thi rd parti S es. BOSCHAllandgFSXgthe
Values, TA=25°C
are regi stered trademarks of1Robert Bosch GmbH, LSB/mg
BNO055: data sheet
SensitNote: Speci fi cati ons wiSthi
ivit y t olerance tol
n thi s document
symbol
are subj
Ta=25°C, ect to change wi thout noti ce.±1
gFS2g ±4
Germany.
%
onboard A/D
Sensit ivit y Temperat ure TCS gFS2g, ±0.03 %/K
Drift Nominal VDD supplies,
Document revision 1.2
Temp operat ing condit ions
Sensit ivit y S VDD gFS2g, TA=25°C, 0.065 0.2 %/V
Document release date
Supply Volt . Drift November 2014
VDD_min ≤ VDD ≤ VDD_max
Zero-g Offset (x,y.z) Offxyz gFS2g, TA=25°C, nominal VDD -150 ±80 + 150 mg Similar offset
Document number BST-BNO055-DS000-12
supplies, over life-t ime
to kionix
Zero-g Offset TCO gFS2g, ±1 + /-3.5 mg/K
Temperat ure Drift Nominal V supplies
Technical reference code(s) 0 273 141 209 DD
Zero-g Offset Supply OffVDD gFS2g, TA=25°C, 1.5 2.5 mg/V

Notes
Volt . Drift
Data in this document are subject to change without notice. Product
VDD_min ≤ VDD ≤ VDD_max
photos and pictures are8 for illustration purposes only and may differ from
nd
Bandwidt h bw8 2 order filt er, bandwidt h Hz
programmable
bw16 the real product’s appearance.
16 Hz
Variable bw 31 31 Hz
Using onboard bw 63 BNO055 63 Hz
Page 14
Filters – choosebw 125 Data sheet 125 Hz
smalles BW bw 250 250 Hz

bw 1,000 Hz
possible for appbwNL
1000
500 500 Hz
Nonlinearit y best fit st raight line, gFS2g 0.5 2 %FS
Out put Noise Densit y nrms gFS2g, TA=25°C 150 190 µg/ Hz
BST-BNO055-DS000-12 | Revi si on 1. 2 | November 2014 Nominal VDD supplies Bosch Sensortec
© Bosch Sensortec GmbH reserves al l ri ghts even i n the event Normal mode
of i ndustri al property ri ghts. We reserve al l ri ghts of di sposal su ch as copyi ng and
passi ng on to thi rd parti es. BOSCH and the symbol are regi stered trademarks of Robert Bosch GmbH, Germany.
Reported as %Full
Note: spec sheets use definitions that give the best numbers!
Note: Speci fi cati ons wi thi n thi s document are subj ect to change wi thout noti ce.
scale output
MECHANICAL C HARACTERISTICS ACCELEROMETER
Cross Axis Sensit ivit y CAS relat ive cont ribut ion bet ween 1 2 %
any t wo of t he t hree axes
Alignment Error EA relat ive t o package out line 0.5 2 °
OPERATING C ONDITIONS GYROSCOPE

Parameter Symbol Condition Min Typ Max Unit
Rat e Range RFS125 Select able 125 °/s

via serial digit al int erface
RFS250 250 °/s
RFS500 500 °/s
RFS1000 1,000 °/s

RFS2000 2,000 °/s
OUTPUT SIGNAL GYROSCOPE

(GYRO ONLY MODE )
Picking sensors
Automotive Applications:
• What might you want to measure for each?
• What sensors might you use?
• How will you use the data?
1. Seat occupancy
2. Airbag deployment
3. Tire pressure monitoring system
4. Cruise control

1. Seat occupancy
2. Airbag deployment

3. Tire pressure monitoring system
4. Cruise control

IoT Circuits
Boris Murmann
[email protected]
The data-driven world
Kim, 2015 International Solid-State Circuits Conference (ISSCC), Keynote Talk

Explosion in wireless connectivity
IoT will be the “next big thing”

Today’s hardware: Bluetooth/ZigBee/WIFI/… modules
Under the hood of an IoT chip
A. Klinefelter et al., "A 6.45µW self-powered IoT SoC with integrated energy-harvesting power management and
ULP asymmetric radios," 2015 IEEE International Solid-State Circuits, pp. 384-385, Feb. 2015.

Generic block diagram

Energy
Memory
Management
Batteries
http://www.silabs.com/whitepapers/battery-life-in-connected-wireless-iot-devices

Batteries improve slowly
§ Semiconductor chip performance tends
to improve rapidly
› See e.g. the blue curve on the left for
microprocessors
› DMIPS = Dhrystone Million
Instructions per Second
§ Battery capacity improves slowly
› About 5-8% per year
› Doubling in ~10 years
§ Key challenge: How to fill this gap via

creative circuit & system design?
Module outline
§ Battery capacity (exercise)

§ Energy management
§ Wireless link
§ Digital computing
§ Analog-digital interfaces

Exercise
§ Estimate the average current that be sustained for a battery life of 10 years
Battery 2 x AA CR2032
Capacity 1000 mAh 225 mAh
Average current
? ?
(for 10-year lifetime)
Solution
§ 10 years = 87,600 hours
Battery 2 x AA CR2032
Capacity 1000 mAh 225 mAh
Average current
11.4 µA 2.6 µA
(for 10-year lifetime)
§ Only a very small amount of current available!

Energy management

Energy
Memory
Management
Typical current consumption across modes
Atmel ATmega2564RFR2 wireless MCU power/sleep mode profiles
§ How to achieve average current consumption of ~10 µA (or less)?
https://www.digikey.com/en/articles/techzone/2016/apr/achieving-wireless-connectivity-for-iot-applications-in-a-power-budget-efficient-way

Key: Duty-cycling
§ IoT transceivers are “mostly” off

§ Example: Receive/transmit within 1 ms, then go to sleep for 1 sec
http://www.mouser.com/applications/low-power-ewc-design/
Battery lifetime calculation becomes complicated….
http://uk.farnell.com/calculating-battery-life-in-iot-applications

Extending battery life (or eliminate batteries!) through harvesting
Most
popular
A. Klinefelter et al., "A 6.45µW self-powered IoT SoC with integrated energy-harvesting power management and
ULP asymmetric radios," 2015 IEEE International Solid-State Circuits, pp. 384-385, Feb. 2015.
How much solar power can we pick up?
Laser Light
§ Typically expect 1 µW/mm2 under
Sunny Day
reasonable lighting conditions
Soccer Stadium § A few cm2 of solar cells may help
Overcast Day
extend the lifetime to “infinity”
Corridor Light
Street Light
Candle Light
10-3 10-2 10-1 100 101 102 103 104 Power (μW/mm2)
0
10 -1
10 10 1
10 2
10 3
10 4
10 5
10 6 Illuminance (Lux)
Chart by Nishit Shah, Stanford

DC-DC converters for energy harvesting
Storage capacitor
§ Want to cover a variety of harvesting sources à Large input voltage range

§ Want relatively stable output voltage
§ Need efficient management of storage capacitor
M. Saadat and B. Murmann, "A 0.6V - 2.4V Input, Fully Integrated Reconfigurable Switched-Capacitor DC-
DC Converter for Energy Harvesting Sensor Tags," in Proc. IEEE Asian Solid-State Circuits Conf., Nov. 2015.
Series-Parallel Converter VCP Load VOUT

VIN
Current
Sensor
CL
VCP
5 nF RL
§ Example design
VDD Current
Selector Controlled
Oscillator R1
§ Switched-capacitor
series-parallel converter
VDDL-IN
POR
provides 8 gain rations

CL
Φ1 R2
Non-overlapping Clock
Switch Control Multiplexer Φ2
Generator
State § Controller adjusts
Machine
CLK
Gain Reset Caps

frequency for optimum
Controller
efficiency
and
Hold Gain
Rectifier
Q
HOLD
3-Bit
Up/Down
Counter ÷N § Provides step charging of
U
R1
Reference
storage capacitance
RREF
VREF PGOOD
R2 Generator Detection
VFB PGOOD
M. Saadat and B. Murmann, "A 0.6V - 2.4V Input, Fully Integrated Reconfigurable Switched-Capacitor DC-DC
Converter for Energy Harvesting Sensor Tags," in Proc. IEEE Asian Solid-State Circuits Conf., Nov. 2015.

Step charging example
𝐶"#$%& = 100𝑛𝐹
Vout (V)
𝑉-. = 0.8𝑉
𝑉$1# = 1.3𝑉 Time
1
Energy loss without step charging: 𝐸4$"" = 𝐶 𝑉 6 = 192 𝑛𝐽
2 "#$%& $1#
1 6
Energy loss with step charging: 𝐸4$"" = : 𝐶"#$%& 𝑉$1#,< − 𝑉$1#,<>? = 18 𝑛𝐽
2
<
M. Saadat and B. Murmann, "A 0.6V - 2.4V Input, Fully Integrated Reconfigurable Switched-
Capacitor DC-DC Converter for Energy Harvesting Sensor Tags," in Proc. IEEE Asian Solid-
State Circuits Conf., Nov. 2015.
Wireless transceiver

Energy
Memory
Management

Power/energy cost of wireless transmission
Bluetooth LE Nike+ ZigBee WIFI

Power (mW) 0.147 0.675 35.7 210
Bits/sec 960 272 192 40M
Energy/bit (nJ) 153 2480 186,000 5.25
§ Numbers are a strong function of reach

§ Energy per bit tends to be lower for high-data rate links à Invested power
amortizes better
https://www.digikey.com/en/articles/techzone/2011/aug/comparing-low-power-wireless-technologies
Pushing the limits

60 GHz pulses
Power
proportional to
data rate
§ A variety of non-standard, short-range radios are being investigated for much

lower energy/bit numbers
§ This open the door for RF powering (no batteries, no solar cells)!
M. Taghivand, K. Aggarwal, Y. Rajavi and A. S. Y. Poon, "An Energy Harvesting 2x60 GHz
Transceiver With Scalable Data Rate of 38–2450 Mb/s for Near-Range Communication," in
IEEE Journal of Solid-State Circuits, vol. 50, no. 8, pp. 1889-1902, Aug. 2015.

RF harvesting circuit
3 nF, on chip
§ Supplies up to 1 mA at 5-10 cm distance
§ RF power can be supplied by cell phone
Comparison of recent works

Ant-sized radio
https://web.stanford.edu/~arbabian/Home/IoT_Radio.html
Digital circuits

Energy
Memory
Management

To transmit or not to transmit?
§ Sending bits is expensive

§ How about adding some
signal processing to reduce
number of bits?
§ For example, how about
data compression?
???
Drawing by Peter Kairouz, Stanford
Typical energy of digital operations (45nm CMOS)
V. Sze, Y.-H. Chen, J. Emer, A. Suleiman, Z. Zhang, “Hardware for Machine Learning: Challenges
and Opportunities,” IEEE Custom Integrated Circuits Conference (CICC), Invited Paper, May 2017.

Energy of image processing accelerators
V. Sze, Y.-H. Chen, J. Emer, A. Suleiman, Z. Zhang, “Hardware for Machine Learning: Challenges and
Opportunities,” IEEE Custom Integrated Circuits Conference (CICC), Invited Paper, May 2017.
Research: Always-on object detection
Feature-space transmission
20x less data

A. Omid-Zohoor, C. Young, D. Ta and B. Murmann, "Towards Always-On Mobile Object
Detection: Energy vs. Performance Tradeoffs for Embedded HOG Feature Extraction," to
appear, IEEE Trans. Circuits and Systems for Video Technology.

Microcontroller power modes
Critical for long

battery life
Core Power Consumption of TI CC2650MODA
http://uk.farnell.com/calculating-battery-life-in-iot-applications
Why can’t we reduce leakage current to zero?
§ MOS transistors turn off gradually

§ Current drops exponentially below threshold
voltage (VT)
§ Two options for reducing leakage

› Increase VT à But this reduces on-current
› Increase subthreshold slope à very difficult;
bounded by 60 mV/decade (kT/q)
VGS

Body biasing in fully depleted silicon-on-insulator (FD-SOI)
FBB = Forward body biasing
RBB = Reverse body biasing
Courtesy Andreia Cathelin, STMicroelectronics
VT changes with body biasing
E. Beigne et al., "Ultra-Wide Voltage Range designs in Fully-Depleted Silicon-On-Insulator FETs," 2013
Design, Automation & Test in Europe Conference & Exhibition (DATE), Grenoble, France, 2013, pp. 613-
618.

Impact on leakage and drive strength (frequency)
Courtesy Philippe Flatresse, STMicroelectronics
Mixed-signal interfaces

Energy
Memory
Management

Anatomy of an analog-digital interface
Key issue for sensors: 1/f noise
offset
+ drift
thermal noise
signal
fcorner log f
§ Analog MOS circuits come with a variety of low frequency artifacts that can
easily ʺswamp outʺ DC or low-frequency signals of interest

Chopper amplification
Analog-to-digital conversion
6
10
Energy per A/D conversion
4
10
P/f s [pJ]
2
10
Flash
Pipeline
0 SAR
10
DS
Other
20 30 40 50 60 70 80 90 100
SNDR [dB] B. Murmann, "ADC Performance Survey 1997-2016," [Online]. Available:
http://web.stanford.edu/~murmann/adcsurvey.html.

The most efficient architecture for IoT applications: SAR ADC
Successive
Approximation
Register
“Sample and Hold”

or
“Track and Hold”
Maloberti, Understanding Microelectronics: A Top-Down Approach, Wiley, 2011
The most efficient architecture for IoT applications: SAR ADC
Maloberti, Understanding Microelectronics: A Top-Down Approach, Wiley, 2011

Implementation example
§ Consumes only 1 nW at 1 kS/s!

› Often negligible compared
to RF and digital energy
§ Conversion occurs within 5 µs

of 1 ms period
› Converter is off/idle most of
the time
P. Harpe, et al. "A 3nW signal-acquisition IC integrating an amplifier with 2.1 NEF and a
1.5fJ/conv-step ADC," 2015 IEEE International Solid-State Circuits Conference, Feb. 2015.
Input amplifier
DC servo loop
§ 32 dB gain
§ AC coupled input
§ 370 Hz bandwidth
§ Input noise = 26 µVrms
§ Total current = 1 nA!
P. Harpe, et al. "A 3nW signal-acquisition IC integrating an amplifier with 2.1 NEF and a
1.5fJ/conv-step ADC," 2015 IEEE International Solid-State Circuits Conference, Feb. 2015.

Summary
§ The IoT has become a significant driver for the semiconductor industry
§ The biggest circuit challenges are related to energy supply
› Batteries won’t improve significantly
› We may not want to deal with batteries in billions of sensors!
§ Take-homes on functional blocks
› Energy management requires handcrafting; must often fuse multiple sources
› RF transmission is most power hungry; must be heavily duty cycled
› Large transmit energy warrants efforts toward increased local computing
› Local compute strongly benefits from optimal supply & VT control
› Analog interfaces are non-trivial (1/f noise, etc.), but not a bottleneck
§ Future applications will decide in which direction we will take the next wave of
circuit & system design innovations
Related Stanford coursework on analog circuits

The Internet of Things
Embedded Systems
Philip Levis
Associate Professor
Computer Science and Electrical Engineering
Stanford University
Embedded Systems
The "Things" in "Internet of Things"
Computing systems designed for a particular application
smart lightbulb
medical implant
aircraft engine controller/sensor
personal fitness tracker
Cost and energy are paramount and greatly define design
Hardware and software often selected together, driven by application

requirements, and software is often highly customized

This module
You'll learn what embedded systems are and how they differ from traditional
computing systems such as servers, phones, and laptops
You'll learn the network and software architectures they typically use and why
You'll learn about the technology drivers behind the recent growth of the IoT
You'll learn about energy budgets and how to design an embedded system
You'll learn about common software techniques and tradeoffs
The Internet of Things


Elicia White
Internet(s) of Things
Industrial Home Area Personal Area Networked

Automation Networks Networks Devices
Thousands/person Hundreds/person Tens/person Tens/person
Controlled environment Uncontrolled environment Personal environment Uncontrolled environment
High reliability Unlicensed spectrum Unlicensed spectrum Unlicensed spectrum
Control networks Convenience Instrumentation Convenience
Industrial requirements Consumer requirements Fashion vs. function Powered
WirelessHART, 802.15.4 ZigBee, ZWave, Thread Bluetooth/BLE WiFi/802.11

6tsch, RPL 6lowpan, Thread 3G/LTE TCP/IP
IEEE/IIC/IETF IETF/ZigBee/Thread/Private 3GPP/IEEE IEEE/IETF

devices
6lowpan, 3G/4G
ZigBee, Ethernet
ZWave,
Bluetooth, TCP/IP
WiFi,
WirelessHART
IoT Systems and Applications

Wide range of applications and application requirements
§ Marketing uses the same term (IoT), but very different
Commonly a three-tier architecture: embedded, gateway, and cloud
Each tier has different computing systems
§ Embedded: microcontrollers, custom software, energy
§ Gateway: often ARM, Android, Linux, iOS
§ Cloud: x86_64, virtualized servers (Linux, etc.)
Application is spread across tiers
§ Software at each tier receives, processes, and displays data

Two Game-Changers
ARM Cortex M series



Example Part: nRF51822
Cortex M0+ with integrated 2.4GHz transceiver

§ Supports Bluetooth Low Energy
§ Two models: 32kB/256kB or 16kB/128kB
DigiKey cost for 25,000: $1.99
6lowpan
We can now inexpensively build powerful networked embedded devices
Early networking: proprietary, vertical silos
§ ZigBee defines link up to application layer
§ Zwave, etc.
§ Connect to Internet through application gateway (lock-in)
6lowpan: IETF standard for IPv6 over low-power link layers
§ Just defines IP packet formats/compression (plus UDP)
§ Routing/MAC layer independent
§ Allows interoperability and software flexiblity
§ RFC4944 (basic protocol) and RFC6282 (additional header compression)

Why Today?
Wireless embedded sensors have been in the market for > 15 years
§ Built around 8-bit or 16-bit microcontrollers, proprietary protocols
Two major technological shifts
§ 32-bit CortexM microcontrollers became efficient enough to use
§ Smartphone support for Bluetooth Low Energy allows mobile devices to interact
with ubiquitous ones
Specification of 6lowpan opened low-power wireless devices ("things") up to the
Internet, hence the "Internet of Things"
Two Game-Changers
ARM Cortex M series



Two Game-Changers
ARM Cortex M series


CortexM microcontrollers
32-bit microcontrollers, use Thumb-1 and Thumb-2 instruction sets

Many types: M0, M0+, M1, M3, M4, M7, M23, M33
§ Optional floating point support: M4F single precision, M7F double precision
Three major subfamilies/architectures
§ ARMv6-M: Cortex-M0, Cortex-M0+, Cortex-M1
§ ARMv7-M: Cortex-M3
§ ARMv7E-M: Cortex-M4, Cortex-M7
§ Upwards compatible
Wikipedia has excellent details on all of the options/features
§ https://en.wikipedia.org/wiki/ARM_Cortex-M

CortexM
CortexM defines an instruction set and core processor functionality
Several pieces of optional functionality
§ Memory protection unit: used to protect firmware from applications
§ SysTick: standard timer
§ Bit banding: spread a word across 32 words, so bits can be individually
read/written: useful for GPIO, other single-bit writes
On-chip features/components modeled as memory-mapped peripherals
§ Each peripheral has a block of addresses for its registers
One huge benefit: compiler/tool support

CortexM has excellent open-source compiler support
§ GCC
§ llvm
§ Leverages importance of general ARM architecture
msp430-gcc recently taken in-house my TI (mspgcc)
avr-gcc is incorporated into Atmel studio
One issue: integrated development environments (AVR studio, code composer

studio) tend to minimize activation energy but eventually become restrictive

CortexM memory map (ARMv6-M)
0xFFFFFFFF
System
0xE0000000
Device not shared
0xC0000000
Device shared
0xA0000000
RAM write-back
0x80000000
RAM write-through
0x60000000
peripherals
0x40000000
SRAM MCUs focus here
0x20000000
Code
0x00000000
CortexM
Every vendor has completely different peripheral interfaces, different chips by
the same vendor can differ significant too
§ CortexM makes it possible to use the same compiler, boot code, interrupt
handling code
§ Everything else needs to be implemented per-chip
§ Switching from one MCU family to another is a lot of engineering effort
§ E.g., switching from Atmel SAM4L to NXP K66

Two Game-Changers
ARM Cortex M series



Not backwards compatible with Bluetooth: additional protocol that sits alongside
traditional Bluetooth
§ Uses similar physical layer (modulation, channel hopping, etc.)
§ Allows transcievers to easily support both
Intended for low data-rate, ubiquitous computing applications
§ Much lower energy consumption
§ Much lower throughput

BLE basics: link layer
BLE nodes can send advertisements (short broadcast packets) freely
§ iBeacon, other beacon technologies use this
Connections use a star topology: a central node can have multiple peripherals, all connections
are between a central and a peripheral
§ Centrals cannot connect to other centrals, peripherals cannot connect to other peripheral,
peripheral can connect to only one central
§ A single device can be both, or alternate: FitBit -> iPhone -> iPad
Central node sets the communication schedule, issues connection request to peripheral: low
power
central
advertiser
peripherals
BLE basics: application layer

Generic ATTributes (GATT) are how BLE devices exchange data
Attributes have a handle, a type, and data
Attributes are characteristics, services, or profiles
§ Form a data hierarchy for querying/discovery
Many standard characteristics, services, profiles
§ Many devices just invent their own
A GATT server provides data to a GATT client
§ Server/client independent of central/peripheral
server client

Overview
CortexM microcontrollers have many sub-families
§ Increasingly powerful instruction sets
§ Common memory layout, interrupt handling, etc.
§ Each vendor has its own peripherals as memory-mapped I/O
§ Leads to excellent compiler support
Bluetooth Low Energy is an ultra-low power link layer
§ Advertisements (beacons) for simple data transmission
§ Central node controls the schedule for low power operation
§ Connections can use GATT, a hierarchical data representation for requesting
particular data
Microcontrollers and Systems-on-Chip (SoCs)

Microcontroller executes code, has on-chip peripherals
§ More powerful costs more, higher power
§ Cortex-M0, CortexM4, Cortex-M4F, …
§ On-chip peripherals (SPI, ADC, CAN
System-on-chip (SoC) combines a microcontroller core with more complex
subsystems/peripherals
§ Radios, Digital signal processors (DSPs)
§ Exact line between "microcontroller" and "SoC" is very blurry (don't worry
about it)
This module: learn about an example MCU and SoC

Atmel SAM4L series: functionality
Cortex M4 core
128-512kB flash (program) memory, 32-64kB RAM
Operates at up to 48MHz at 1.68 to 3.6V
4 USART (UART or SPI) busesm 4 I2C (two-wire) buses
Hardware AES encryption support
USB hardware support
8-bit or 12-bit ADC (3-15 channels depending on form factor)
15 DMA channels for I/O processing offload
1.5-3µA sleep, 1.5µs wakeup, as low as 90µA/MHz active (4.3mA @ 48MHz)
http://www.atmel.com/products/microcontrollers/ARM/SAM4L.aspx
Nordic nRF51 SoC

Cortex-M0 core
Integrated Bluetooth Low Energy transciever
Operates at 16MHz at 1.68 to 3.6V
1 UART, 1 SPI, one I2C bus
Hardware AES encryption support (limited)
8-bit, 9-bit, 10-bit ADC
2.6µAsleep, 4.2µs wakeup, 2.4-4.1mA active (RAM vs. flash)
16mA TX (@ +4dBm), 13.4mA RX (but radio is usually off)

Takeaways
Huge variations in what different MCUs/SoCs can do
Orders of magnitude differences in active vs. sleep current
RF is much more expensive than compute
Two basic cost considerations dominate

Energy
§ Embedded systems have an expected workload – want to minimize energy
required to handle the workload (longer battery life, more efficient)
§ More powerful, highly featured MCUs draw more power: can be more power
efficient, but consume more energy for a given workload
Money
§ Many embedded systems markets have tight margins (cars, appliances, etc.)
§ More powerful, highly featured MCUs cost more
Both costs push designs to pick minimal MCU and optimize code for it

Energy examples
Cortex M4 core
Operates at up to 48MHz at 1.68 to 3.6V
4 USART (UART or SPI) busesm 4 I2C (two-wire) buses
Hardware AES encryption support
USB hardware support
8-bit or 12-bit ADC (3-15 channels depending on form factor)
15 DMA channels for I/O processing offload
1.5-3µA sleep, 1.5µs wakeup, as low as 90µA/MHz active (4.3mA @ 48MHz)
http://www.atmel.com/products/microcontrollers/ARM/SAM4L.aspx
Nordic nRF51 SoC

Cortex-M0 core
Integrated Bluetooth Low Energy transciever
Operates at 16MHz at 1.68 to 3.6V
1 UART, 1 SPI, one I2C bus
Hardware AES encryption support (limited)
8-bit, 9-bit, 10-bit ADC
2.6µAsleep, 4.2µs wakeup, 2.4-4.1mA active (RAM vs. flash)
16mA TX (@ +4dBm), 13.4mA RX (but radio is usually off)

Cost examples
Part Family Flash RAM Cost Notes
ATSAMD20E15 Cortex-M0+ 16kB 2kB $1.37
A
ATSAMD21E16 Cortex-M0+ 64kB 8kB $1.70 LIN, USB
B
NRF51422 Cortex-M0 256kB 32kB $2.44 BLE
ATSAMD20J18A Cortex-M0+ 256kB 32kB $2.69 20 12-bit ADC
ATSAM4S2BA Cortex-M4 128kB 64kB $2.80 SSC, USB
ATSAMD21G18 Cortex-M0+ 256kB 32kB $3.15 LIN, USB
A Data from DigiKey, July 2017, for ~1,000 units ; at scale costs are much lower
ATSAM4E8EA Cortex-M4 512kB 128kB $7.62 CAN, Ethernet, USB, IrDA
Takeaways
If energy is important, sleep as much as possible and avoid using radio
§ Orders of magnitude differences in active vs. sleep current
§ RF is much more expensive than compute
Choose the smallest MCU you need for the application
§ Huge variations in what different MCUs/SoCs can do
§ Cost increases with features

Energy is the limiting resource
Many embedded systems are battery-powered
§ Consuming less energy means longer lifetime or fewer recharges
Calculating an energy budget allows you to reason about design tradeoffs
E: energy
E = P s ⋅ ts + P a ⋅ ta P: power
t: time
E = sleep energy + active energy s: sleep
a: active
Example energy budget

nRF51422 that wakes up at 1Hz and transmits a single BLE advertisement
E = P s ⋅ ts + P a ⋅ ta E: energy
P: power
E = sleep power * sleep time + active power * active time t: time
s: sleep
E = 4.8µA * sleep time + 14.6mA * active time
a: active
E = 4.8µA * 999.7ms + 14.6mA * 0.3ms
E = 4.8uAs + 4.38µAs
CR2032 is 225mAh – in theory can last 3 years (in practice much less)
Sleep current assumes all RAM retained; active current assumes 0dB transmit plus CPU at 16MHz
Active time is 37 bytes @ 1Mbps = 300µS

Alarm bells should be going off
Energy is power * time, not current * time
Often use current when calculating microcontroller power budgets
§ Batteries do not provide constant voltage
§ Don't use voltage regulators because of significant (energy) costs
§ Take advantage of MCU tolerances (e.g. 1.8-3.6V)
Complication
We assumed that system instantaneously transitioned from sleep to wake
§ MCU takes time to wake up
§ Transceiver takes time to power up
Transition times – high power but no work – can be significant
§ Wake up less often, for longer, to amortize over wake period
E = P s ⋅ ts + P a ⋅ ta + P T ⋅ tT
E = sleep energy + active energy + transition energy

Example
It takes the nRF51422 140µS to transition to TX state, draws 7mA
E = P s ⋅ ts + P a ⋅ ta + P T ⋅ tT
E: energy
P: power
E = 4.8µA * 999.6ms + 14.6mA * 0.3ms + 7.0mA * 0.14ms t: time
s: sleep
E= 4.8uAs + 4.38µAs + 1µAs
a: active
T: transition
Not considering transition times is off by 10%
Takeaways
Energy consumed is the sum of active and sleep energy
For ultra-low power applications, sleep and active energy can be nearly equal
Transition costs can also be significant
Based on energy budget, pick battery with desired capacity

Minimizing Energy Consumption
To minimize sleep energy, put microcontroller into lowest possible state
§ Microcontrollers have different many low-power states and power-saving
features
§ Complex implications to software
To minimize active energy, minimize time peripherals and MCU are active
§ Perform operations in parallel to minimize active time
§ Cluster/batch operations to minimize transition times
§ Minimize clock rate
Sleep states: SAM4L

Four basic running modes
§ Run – MCU executes instructions, everything can be active
§ Sleep – no instructions, any clocks/peripherals can be active
§ Wait – no instructions, only 32kHz clock active for peripherals
§ Retention – no instructions, only 32kHz clock, no peripherals
Mode Current Wakeup latency
Run (@48MHz) 14.5mA -
Sleep 50µA 0.25µs
Wait 6µA 1.5µs
Retention 3µA 1.5µs

Sleep state: SAM4L clocks
SAM4L has many different clocks: 32khZ, 1, 4, 8, 12, 80MHz
Higher speeds draw more power: use slowest clock possible (dividers)
But, if a fast enough clock is already on, can subdivide for slower rates
E.g., if need 1MHz and If RCFAST is on, use a clock divider of 12

Clock Speed Current
OSC32K 32kHz 350nA
RCSYS 116kHz 2µA
RC1M 1MHz 35µA
RCFAST 12MHz 180µA
Parallelism
Parallelism allows fixed overheads to amortize over multiple operations
loop {
loop {
sample_sensor();
parallel {
radio_on();
sample_sensor();
send_value();
radio_on();
sleep();
}
}
send_value();
High-cost clocks are on longer, sleep();
in SLEEP (50µA) state longer. }
When could this be more efficient?

Batching: LTE as an example
LTE Power States
data packet
data packet 100ms 20ms

High Low
Idle On Duty
(RRC_Idle, (Continuous Duty
0.33%) Reception) (Short DRX, (Long DRX,
data packet 5% duty cycle) 2.5% duty cycle)
data packet
11.5s
INdT/Microsoft
Power Management in Mobile Devices (2014) 4 Philip Levis, Stanford University
LTE Batching
Using LTE, compare

§ Sending one small packet every 30s
§ Sending two small packets every 60s
One packet every 30s

§ (0.1s * 100%) + (0.02 * 5%) + (11.5 * 2.5%) + ((30-11.62) * 0.33%)
§ 0.1 + 0.001 + .2875 + .06 = .3576s/30s
§ = 1.2%
Two packets every 60s
§ (0.101s * 100%) + (0.02s * 5%) + (11.5s * 2.5%) + ((60-11.62)s * 0.33%)
§ 0.101 + 0.001 + .2875 + .160 = .4577s/60s
§ = 0.76%

Minimize clock rate
Most peripherals draw current based on clock rate
§ UART: 8.5µA/MHz
§ SPI: 1.9µA/MHz
Lowering the clock rate reduces power drawn; if a fast enough clock is already
on, can subdivide for slower rates
E.g., if need 1MHz and If RCFAST is on, use
Clock Speed Current
a clock divider of 12
O0SC32K 32kHz 350nA
RCSYS 116kHz 2µA
RC1M 1MHz 35µA
RCFAST 12MHz 180µA

features

CortexM
CortexM memory map (ARMv6-M, ARMv7-M)

0xFFFFFFFF
System
0xE0000000
Device not shared
0xC0000000
Device shared
0xA0000000
RAM write-back
0x80000000
RAM write-through
0x60000000
peripherals
0x40000000
0x20000000
Code
0x00000000

Core registers r0 return value
r1
r2 function arguments
r3
r4
r5
r6
r7
r8
r9
r10
r11 (fp) frame pointer
r12
r13 (sp) stack pointer
r14 (lr) link register
r15 (pc) program counter
Core registers
Handler Thread
r0 r0 return value
r1 r1
r2 r2 function arguments
r3 r3
r4 r4
r5 r5
r6 r6 Two operating modes:
r7 r7
r8 r8
thread and handler
r9 r9
r10 r10
r11 (fp) r11 (fp) frame pointer
r12 r12
r13 (sp) r13 (sp) stack pointer
r14 (lr) r14 (lr) link register
r15 (pc) r15 (pc) program counter

Control Registers (M4)
r0 return value
r1
r2 function arguments
r3
r4
r5 APSR N Z C V Q
r6
r7 IPSR 0 or Exception
r8 ICI ICI
r9 EPSR T
/IT /IT
r10
r11 (fp) frame pointer
r12
r13 (sp) stack pointer
r14 (lr) link register
r15 (pc) program counter
Interrupt Vector table (M4)

At address 0x0 (can be relocated with VTOR register)
IRQ1 chip specific/defined

IRQ0 0x0044
SVC
0x002C
reserved
0x001C
usage fault architecture defined
bus fault
mem fault
hard fault
NMI
reset
0x0000

CortexM
CortexM memory map (ARMv6-M, ARMv7-M)

0xFFFFFFFF
System
0xE0000000
Device not shared
0xC0000000
Device shared
0xA0000000
RAM write-back
0x80000000
RAM write-through
0x60000000
peripherals
0x40000000
0x20000000
Code
0x00000000

Peripherals
Each peripheral is mapped to a region (0x1000 in size) of memory
Registers laid out in that memory (completely up to chip)
Most associated with an interrupt (NVIC index/IRQ#)
Offset Register Name Access Reset

SAM4L AESA 0x00 Control CTRL RW 0x00000000
0x04 Mode MODE RW 0x000F0000
AES accelerator 0x08 Data buffer pointer DATABUFPTR RW 0x00000000
§ AES-128 0x0C Status SR R 0x00010000

0x10 Interrupt enable IER W 0x00000000
§ ECB
0x14 Interrupt disable IDR W 0x00000000
§ CBC 0x18 Interrupt mask IMR R 0x00000000
§ CF 0x20 Key 0 KEY0 W 0x00000000
§ OF 0x3C Key 7 KEY7 W 0x00000000
§ CTR 0x40 Initialization vector 0 INITVECT0 W 0x00000000

0x4C Initialization vector 3 INITVECT3 W 0x00000000
0x50 Input data IDATA W -
0X60 Output data ODATA R -
0x70 DRNG register DRNGSEED W 0x00000000

General operation
Configure peripheral for operation (e.g., AES input data, key)
Set bit in control register to start operation
Return data on completion
§ Spin loop on interrupt mask for very fast operations
§ Handle interrupt for longer operations
§ Read data out of data register (if input)
Buses
SCL
I2C SDATA
UAR RX
T TX
CLK
MOSI
SPI MISO
CSn

Larger-scale I/O
Some operations transfer a small amount of data
§ Single byte UART write
§ Single ADC sample
Others transfer more data, sometimes very fast
§ Transfer USB data at 8MHz
§ High frequency ADC sampling (100 of kHz)
For larger transfers, per-byte interrupts can be expensive
§ Wakeup, active power cost energy
§ Software may not handle interrupts fast enough
DMA: Direct Memory Access

Set up a series of operations as a direct memory access operation
§ Provide a buffer to read into and/or write from
§ Tell system how many operations
§ Start operation
Receive DMA interrupt when whole series of operations complete
Amortize interrupt cost across many operations
§ Consumes less energy (fewer CPU cycles)
§ Can run faster (not limited by interrupt rate)

Peripherals
On-chip accelerators for common operations
§ Huge variety, depending on application focus on MCU
Interact/use with memory-mapped registers
Larger operations can use DMA to save energy
Fast operations can use DMA to run faster
Handling Interrupts
Handler Thread
r0 r0 return value
r1 r1
r2 r2 function arguments
r3 r3
r4 r4
r5 r5
r6 r6 Two operating modes:
r7 r7
r8 r8
thread and handler
r9 r9
r10 r10
r11 (fp) r11 (fp) frame pointer
r12 r12
r13 (sp) r13 (sp) stack pointer
r14 (lr) r14 (lr) link register
r15 (pc) r15 (pc) program counter

Execution model
Thread mode executes main loop (application code, etc.)

Handler mode executes interrupts
§ Interrupts can happen at any time
§ Can be disabled, but doing this for a long time is dangerous
interrupts
application application
The danger of interrupts

Thread mode Handler mode
extern int a; extern int a;
void inc() { void dec() {
a = a + 1; a = a - 1;
} }
00008000 <inc>: 0000802c <dec>:

8000: e52db004 push {fp} 802c: e52db004 push {fp}
8004: e28db000 add fp, sp, #0 8030: e28db000 add fp, sp, #0
8008: e59f3018 ldr r3, [pc, #24] 8034: e59f3018 ldr r3, [pc, #24]
800c: e5933000 ldr r3, [r3] 8038: e5933000 ldr r3, [r3]
8010: e2832001 add r2, r3, #1 803c: e2432001 sub r2, r3, #1
8014: e59f300c ldr r3, [pc, #12] 8040: e59f300c ldr r3, [pc, #12]
8018: e5832000 str r2, [r3] 8044: e5832000 str r2, [r3]
801c: e24bd000 sub sp, fp, #0 8048: e24bd000 sub sp, fp, #0
8020: e49db004 pop {fp} 804c: e49db004 pop {fp}
8024: e12fff1e bx lr 8050: e12fff1e bx lr
8028: 00010070 .word 0x00010070 8054: 00010070 .word 0x00010070

The danger of interrupts
Thread mode Handler mode
extern int a; extern int a;
void inc() { void dec() {
a = a + 1; a = a - 1;
} }
00008000 <inc>: 0000802c <dec>:

8000: e52db004 push {fp} 802c: e52db004 push {fp}
8004: e28db000 add fp, sp, #0 8030: e28db000 add fp, sp, #0
8008: e59f3018 ldr r3, [pc, #24] 8034: e59f3018 ldr r3, [pc, #24]
800c: e5933000 ldr r3, [r3] 8038: e5933000 ldr r3, [r3]
8010: e2832001 add r2, r3, #1 danger 803c: e2432001 sub r2, r3, #1
8014: e59f300c ldr r3, [pc, #12] 8040: e59f300c ldr r3, [pc, #12]
8018: e5832000 str r2, [r3] 8044: e5832000 str r2, [r3]
801c: e24bd000 sub sp, fp, #0 8048: e24bd000 sub sp, fp, #0
8020: e49db004 pop {fp} 804c: e49db004 pop {fp}
8024: e12fff1e bx lr 8050: e12fff1e bx lr
8028: 00010070 .word 0x00010070 8054: 00010070 .word 0x00010070
If dec() runs when inc() is at 0x8010, then result of dec() will be lost: it
has copied a into r3.
Dealing with interrupts

Option 1: spin loop to wait (check pending flag)
§ Pro: no latency, no race conditions
§ Con: wasted cycles, no concurrency
Option 2: disable interrupts in critical sections
§ Pro: only small latency, full concurrency
§ Con: critical sections must be small, easy to make mistakes
Option 3: minimize code in interrupt (delay processing to sequential code)
§ Pro: simple code, full concurrency
§ Con: interrupts can be delayed significantly

Software Architecture
applications
libraries services
operating system
hardware
Operating System
Kernel: lowest level of software
§ Directly accesses hardware
§ Has complete control of system
§ Example: Linux, Windows, OSX
Libraries and services on top of kernel
§ Networking, storage, other utilities
General goal: minimize what's in the kernel
§ Challenge: crossing between kernel and libraries is expensive
§ If abstract requires a lot of crossings, move inside kernel

Embedded Operating Systems
Many MCUs do not distinguish kernel and user code
§ No hardware memory protection
§ Application code can crash entire system
§ Requires diligence and care in software design
Trend is changing, with CortexM memory protection unit (MPU)
§ Embedded Oses haven't caught up
Embedded OSes have to walk a fine line
§ Provide useful software abstractions for applications
§ Allow applications to break through abstractions when necessary
Key OS responsibility: execution model

How do you write code that responds to interrupts?
How do you write code that runs in the background?
When can code preempt other code?

Interrupt code
Code in interrupt handler can block entire system (no interrupts handled)
§ Preemptive interrupts are especially challenging (re-use of handler context)
§ Goal: minimize code in direct handler
Use deferred procedure call (interrupt bottom half)
§ Interrupt handler does core work (e.g., pull data out of register into queue)
§ Deferred procedure call executes after interrupt handler returns
§ Interrupts can preempt deferred procedure call/bottom half
Interrupt code
interrupts
application application

Interrupt code
a b c interrupts
application a b c application
bottom halves
Background code
Application code: preempted by interrupts and bottom halves
Less time critical (although may still have deadlines)
Option 1: event-driven
§ Function that executes in response to an event (timer, packet reception)
§ Cross-event state stored in global variables
Option 2: threaded
§ Function that never returns: calls functions that block (resume thread on
completion)
§ Cross-event state stored on stack (higher RAM cost)

Example: event-driven
receive_callback(receive_cb);
#define MAX_LEN 128

uint8_t send_buffer[MAX_LEN];
void receive_cb(void* buf, uint8_t len) {

if (!busy) {
uint8_t safe_len= (len > MAX_LEN)? MAX_LEN: len);
memcpy(buf, send_buffer, safe_len);
send(send_buffer, safe_len);
busy = true;
}
}
Example: threaded
#define MAX_LEN 128
void main() {
while (1) {
uint8_t buffer[MAX_LEN];
len = receive(buffer, MAX_LEN);
send(buffer, len);
}
}
Note: packets received while sending are dropped

FreeRTOS
Intended for real-time use (has priorities, deadlines, etc.)
Thin loop on top of hardware events
Extremely flexible, supports many chipsets
Allows (expects) you to write your own interrupt handlers
Also supports threads (called "tasks")
§ Every task has a priority, higher is more important (idle task is 0)
§ Tasks scheduled preemptively
§ Interrupts can make tasks runnable
ARM Mbed OS
ARM Mbed OS provides
§ Standard APIs for chip peripherals (SPI, I2C, etc.)
§ Libraries on top of standard APIs
Supports both event-driven and threaded programming
§ Threaded programming has easier power management
§ Can write interrupt handlers, etc.
Generally speaking, APIs are richer and cleaner than FreeRTOS
But, actual Mbed implementations are by vendors so be careful

Operating Systems Review
Lowest level of software for applications to build on
Define concurrency model, APIs
Event-driven vs. threaded execution
This module
You've learned what embedded systems are and how they differ from traditional
computing systems such as servers, phones, and laptops
You've learned the network and software architectures they typically use and why
You've learned about the technology drivers behind the recent growth of the IoT
You've learned about energy budgets and how to design an embedded system
You've learned about common software techniques and tradeoffs


Elicia White

devices
6lowpan, 3G/4G
ZigBee, Ethernet
ZWave,
Bluetooth, TCP/IP
WiFi,
WirelessHART

Two Game-Changers
ARM Cortex M series


Computing an energy budget
Many embedded systems are battery-powered

§ Consuming less energy means longer lifetime or fewer recharges
Calculating an energy budget allows you to reason about design tradeoffs
Transition times – high power but no work – can be significant
§ Wake up less often, for longer, to amortize over wake period
E = P s ⋅ ts + P a ⋅ ta + P T ⋅ tT
E = sleep energy + active energy + transition energy

Software Architecture
applications
libraries services
operating system
hardware

features

IoT Connectivity and Networking
Ayfer Özgür
Stanford University
Historical Perspective on Wireless Systems

• First Wireless Age: Station to Station (Kilo Scale)

• Second Wireless Age: Station to People (Mega Scale)

• Second Wireless Age: Station to People (Mega Scale)
• Third Wireless Age: People to People (Giga Scale)

1000x increase in connectivity in each cycle.
Connecting various devices and sensors to each other and the Internet.

Paradigm shift in principles, technology, systems and applications.
Cost, size, maintenance-free operation critical for scalability.

Eliminating the Battery
Batteries are bulky and relatively expensive, require maintenance.
Energy Harvesting Remotely Powered Devices

• New deployment models and applications.
• True mobility.
• Maintenance-free perpetual operation.
Source:wonderfulenginneering.com

Communication Under Random Energy Dynamics
energy in energy out
• Energy: • Energy:
• Static • Dynamic
• Deterministic • Random
Communication system design under random energy dynamics.
Massive Multiple Access Networks

Today’s Wireless Systems:
• Moderate system size.
• Large payloads.
• Tightly managed and
synchronized.
• Node identity and message.

Massive Multiple Access Networks
Tomorrow’s IoT Networks:
• Massive number of devices, only
few active at a time.
• Bursty traffic, few bits to transmit.
• Complexity and energy
constrained.
• Node identity may not be relevant.
Source:neoswarm.com
Module Outline
• Energy Harvesting Transmitters for IoT
• Wirelessly Powered Transmitters for IoT
• Massive Multiple Access
• Related Coursework
• Quiz

Basic Model for Communication
Capacity of the AWGN Channel

The most famous formula (Shannon)

Capacity of the AWGN Channel
The most famous formula (Shannon)
Operating Regimes:
W SNR ≪ 0 dB Bandwidth-limited
C∝
P/N0 SNR ≫ 0 dB Power-limited
Communication system design is dictated by the operating regime.
Communication Under Random Energy Dynamics
Et: stochastic process known causally at the transmitter and not at the receiver.

Bmax = ∞
Capacity equal to that of a classical AWGN channel with
Bmax = ∞
Capacity equal to that of a classical AWGN channel with
Insights:
• Use standard communication and coding techniques for the AWGN channel.
• Only relevant property in determining capacity is the average energy harvesting rate.
Ref: O. Ozel, S. Ulukus, Achieving AWGN capacity under stochastic energy harvesting, IEEE
Transactions on Information theory, 2012.

Bmax finite
Foremost engineering questions:
• How does the capacity of the energy harvesting AWGN channel
depend on system parameters such as Bmax and Et?
• What are the properties of Et most relevant to capacity? What are
more favorable and less favorable energy profiles?
• Are there different operating regimes where the dependence to Bmax
and Et is qualitatively different?
• For a given Et , how can we “optimally” choose Bmax?
• What is the optimal power control policy?
Capacity for an Energy Harvesting Transmitter
Ref: D. Shaviv, P.M. Nguyen, A. Ozgur, Capacity of the Energy Harvesting Channel with a
Finite Battery, IEEE Transactions on Information theory, 2016.

Capacity for an Energy Harvesting Transmitter
Ref: D. Shaviv, P.M. Nguyen, A. Ozgur, Capacity of the Energy Harvesting Channel with a
Finite Battery, IEEE Transactions on Information theory, 2016.
Correlations in Energy Arrivals
Capacity decreases with coherence time T in general.

Remotely Powered Communication
Two Topologies for Home IoT
Current practice:
• Transfer energy at a constant rate.
• Periodically charge transmitter’s battery.

Exploit Side Information
Charger observes the output of the Charger observes the input to the
channel. channel.
Binary Example
• Charger has no side information:
• Et = 1, ∀t: C∅ = 1 bits/channel use, Γ = 1 unit/channel use.

Binary Example
• Charger knows the message:
Ø Charge when the transmitter intends to send a 1:
• CM = 1 bits/channel use, Γ = 1/2 units/channel use.
•
Binary Example
• Charger knows the message:
Ø Charge when the transmitter intends to send a 1:
• CM = 1 bits/channel use, Γ = 1/2 units/channel use.
• Charger can observe the transmitted signal Xt−1:
Ø Charge when battery is empty:
• CX = 1 bits/channel use, Γ = 1/2 units/channel use.

Remotely Powered Communications
• Charger: Dynamically decide how much energy to transfer to the receiver based on its side
information regarding the transmission (subject to an average power constraint Γ).
• Transmitter: Dynamically adapt its transmission scheme based on its instantaneous battery level.
Remotely Powered Communications
Exploiting side information at the charger can enable performance close to the centralized case.
Ref: D. Shaviv, A. Ozgur, H. Permuter, Capacity of Remotely Powered Communication, IEEE Transactions on Information theory, 2016.

Reservation Systems
frequency and time resources are
tightly allocated across active devices
• LTE, GSM etc.
• Narrowband IoT
Overhead for coordination becomes the bottleneck when the number of

devices is large and payload is small.
Random Access Systems

CSMA, CDMA, Aloha Wifi, Lora, ZigBee, BLE, etc.
• Carrier-sensing is energy-consuming.
• Aloha leads to high collision rates.

Embracing Collisions
Design codes that are resilient up to a given number of collisions.
Node Identification
• On-off keying used for transmission.

• Energy detection at the receiver.
• At most d out of N users active at a time where d ≪ N.

Node Identification
Trivial Solution: t = N.
Group Testing
• N items (soldiers).
• d defective (infected items).
• Figure out the defective group.
• Group items together and apply t tests.

Group Testing
Group testing literature:
t = Θ(d log N)
Collision Resolving Codes via Group Testing
Node identification: assign a single distinct column to each users.

Collision Resolving Codes vis Group Testing
Joint Identification and Information Transmission: assign multiple distinct columns to each user.
Communication vs Computation
How to trade-off communication vs computation.

Communication vs Computation
Compute and communicate at multiple stages.
Related Coursework on Communication

XEE100 All Slides

Uploaded by

Copyright:

Available Formats

XEE100 All Slides

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

XEE100 All Slides

Uploaded by

Copyright:

Available Formats

IoT Intro

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Tsai, 2014 International Solid-State Circuits Conference (ISSCC), Keynote Talk

What is IoT? Cool

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Cool Applications – Professor Ada Poon

• The purpose of this section is two fold:

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

• Go from collecting 840

Smart Street Lights

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Connected Vehicles – Combine private and collective

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Enhance Adherence – Ingestible Sensors

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

• Wifi or cellular connectivity

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Capacitive measurement of displacement

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

MEMS sensors: Small is beautiful

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Generic block diagram

Wireless Accelerators &

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

§ IoT transceivers are “mostly” off

Power/energy cost of wireless transmission

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

What is an embedded system?

“An embedded system is a computerized system that

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

IoT: MGC Architecture

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

ARM Cortex M series

Bluetooth Low Energy

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Professor Ada Poon

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Smart Waste Management

Dublin Airport New York City

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Smart Waste Management

• Go from collecting 840

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Smart Street Lights

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Smart Street Parking

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Security without Surveillance

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Baby Monitoring – Activity Tracking

Copyright ⓒ 2017 Stanford University. ALL RIGHTS RESERVED

Baby Monitoring – Activity Tracking

1. The turtle sends information about