IoT: Architecture

and Devices

1
 Outline
1. What is IoT?

2. Architecture of IoT devices

- Sensing subsystem
- Computing subsystem
- Actuation subsystem

M. Poncino - Politecnico di Torino 2

What is IoT?

3
 What is IoT?
• The Internet-of-Things can be seen from many perspectives
- Hardware, software, network, application domain…

• Popular term even among non-experts!

• Let’s see some definitions

M. Poncino - Politecnico di Torino 4

 What is IoT?
[borrowed from
M. Zennaro, https://www.itu.int]

• “Semantically” correct!
- Emphasis on the connectivity (INTERNET!)
- Connected objects are abstract (THINGS)
M. Poncino - Politecnico di Torino 5
 In what it is new or different?
• In what IoT differ w.r.t. Machine-to-Machine (M2M)
interactions
- E.g., a device controlling a timer, a heater, a machine
• Key is the INTERNET!
M2M IoT
Point-to-point communication embedded with the HW Devices communicate via IP networks, incorporating
multiple protocols
Devices use cellular or other Data delivery relayed through a middle-layer hosted in the
ad-hoc networks cloud
Devices do not rely on the internet Devices DO require an internet connection

Integration is limited (devices must match the Virtually unlimited integration (but needs a solution to
corresponding standards) manage all types of communications
M. Poncino - Politecnico di Torino 6
 The IoT hierachy
• A functionality-oriented view

M. Poncino - Politecnico di Torino 7

 The IoT hierachy: Edge, Fog, Cloud
• A device-oriented view

M. Poncino - Politecnico di Torino 8

 Edge, fog, cloud computing
• Cloud Layer: Industrial big data, business logic and analytics
databases and data “warehousing”

• Fog Layer: Local network assets, micro-data centres

• Edge Layer: Real-time data processing on industrial PCs,

process-specific applications and autonomous equipment

M. Poncino - Politecnico di Torino 9

Edge, fog, cloud: Data processing

[Source: https://www.winsystems.com]
M. Poncino - Politecnico di Torino 10
 IoT and this course
• IoT as a whole is a sort of distributed computing system
- Learning IoT cannot be a single course
• Sensors, devices, architectures
• Networks and ad-hoc protocols
• Cloud
• Data management
• Security
• Business models….
• Our focus is on the devices and their energy balance

M. Poncino - Politecnico di Torino 11

 IoT and this course
• IoT as a whole is a sort of distributed computing system
- Learning IoT cannot be a single course
• Sensors, devices, architectures
• Networks and ad-hoc protocols
• Cloud
• Data management
• Security
• Business models….
• Our focus is on the devices and their energy balance
- The “edge devices”
- Understand the architectures, the design issues in particular from the
perspective of the energy consumption
M. Poncino - Politecnico di Torino 12
IoT: Architectures and
Devices
Prof. M. Poncino
DAUIN
IoT device architecture
Source: ti.com

M. Poncino - Politecnico di Torino 14

 A more systematic view

(includes
(filter, amplif.)
network I/F)
sensors Signal Digital actua
sensors ADC DAC actuat
sensors cond. processing actuat

environment

M. Poncino - Politecnico di Torino 15

 A more systematic view

sensors Signal Digital actua

sensors ADC DAC actuat
sensors cond. processing actuat

Sensing subsystem
Processing subsystem
Actuation (communication) subsystem

M. Poncino - Politecnico di Torino 16

 (Extra) Functionality
• This is a functional view
- What the system does sensors
sensors
sensors
Signal
cond.
ADC
Digital
processing
DAC actua
actuat
actuat
in terms of functionality
- Connections represent
flow of data
• What is missing?
• A Non-functional view
- Most important non-functional information is
the flow of energy/power
- Implies adding other components

M. Poncino - Politecnico di Torino 17

 Functional and energy/power flows
• Not just functionality (i.e., pure consumption)…
environment
environment
sensors Signal Digital actua
sensors ADC DAC actuat
sensors cond. processing actuat

conv conv conv conv conv conv

MPPT
scavengers eEnergy Functionality flow
scavengers conv
scavengers Power/energy flow
storage

M. Poncino - Politecnico di Torino 18

 Functional and energy/power flows
• Not just functionality (i.e., pure consumption)…
environment
sensors Signal LOADS
ADC
Digital
DAC actua
environment
sensors
sensors cond. processing actuat
actuat

conv conv conv conv conv conv

MPPT
scavengers eEnergy Functionality flow
scavengers conv
scavengers Power/energy flow
storage
Functional components are generic ‘loads’ from the power flow perspective!!!
M. Poncino - Politecnico di Torino 19
 Lecture plan
• We will first quickly overview the main concepts relative from
the functional perspective
- The sensing subsystem
- The processing subsystem
- The actuation subsystem
• We will then move to the power/energy perspective
- Understand the issues involving energy
• Consuming (as little as possible) energy
• Generating energy
• Storing energy
• Convert energy

M. Poncino - Politecnico di Torino 20

 A more systematic view

sensors Signal Digital actua

sensors ADC DAC actuat
sensors cond. processing actuat

Sensing subsystem
Processing subsystem
Actuation (communication) subsystem
21
M. Poncino - Politecnico di Torino
The sensing sub-system

22
 Introduction
• Sensors are ubiquitous

• Unthinkable to have a system of any complexity without any

sensor!
• Key for “smartness”

M. Poncino - Politecnico di Torino 23

 Active vs. passive sensors
• Active = require an external power supply to operate, called
an excitation signal which is used by the sensor to produce the
output signal.
- Example: Radar/Lidar, IR sensors*

• Passive = does not need any additional power source or

excitation voltage.
- They generates an output signal in response to some external stimulus.
- Example: a thermocouple which generates its own voltage output when
exposed to heat.

M. Poncino - Politecnico di Torino 24

 Analog vs. Digital Sensors

• Analog: produce a continuous output signal or voltage which is

generally proportional to the quantity being measured (most
common)

• Digital: produce a discrete digital output signals or voltages

that are a digital representation of the quantity being
measured.

M. Poncino - Politecnico di Torino 25

 Analog vs. Digital Sensors
• Analog: Thermocouple used to produce an Analog Signal

[Source: electronics-tutorial.com]

M. Poncino - Politecnico di Torino 26

 Analog vs. Digital Sensors
• Digital: Light Sensor used to produce a Digital Signal

[Source: electronics-tutorial.com]

M. Poncino - Politecnico di Torino 27

 Sensing what?
• Basically everything (see list on Wikipedia…)
- Acoustic, sound, vibration
- Chemical (includes all biomedical ones)
- Electric current, electric potential, magnetic, radio
- Environmental
- Flow, fluid velocity
- Ionizing radiation, subatomic particles
- Navigation
- Position, angle, displacement, distance, speed, acceleration
- Optical, light, imaging, photon
- Pressure, Force, density, level
- Thermal, heat, temperature
- Proximity, presence
M. Poncino - Politecnico di Torino 28
 Some examples
• A simple one: a microphone
- Can be very technological (e.g., a MEMS microphone)

M. Poncino - Politecnico di Torino 29

 Some examples
• A more sophisticated one: PM 2.5-10 sensor
- E.g., for smoke, dust, pollen,
- Works by detecting with a photodiode
the light scattered by particles
suspended in air in a chamber
- A circuit generates a series of pulses
TIDA-00378 PM2.5 and PM10
that are directly related
to the light scattered
by the particles
- Needs a lot of
processing!

M. Poncino - Politecnico di Torino 30

 What exactly is a ‘sensor’?
• Key element is the transducer
• Transducer = element that converts energy from one domain
to another
energy Xducer (electrical)
energy

- Typically we are interested into transducers that have electrical domain

as output
• A sensor (as well as an actuator) includes a transducer
- Sometimes the same actuator con serve in both directions (e.g., an
antenna)
M. Poncino - Politecnico di Torino 31
 What exactly is a ‘sensor’?
• Overall conceptual structure
electrical
signal
Input signal
Signal
Xducer Conditioning

SENSOR

M. Poncino - Politecnico di Torino 32

 Signal Conditioning
• Especially for analog sensors, their output cannot
be used “as is”
• Some conditioning is necessary!

- Signal-level and bias changes

• adjust the level (magnitude) and bias (zero value) of signal to
match the desired levels
• Example:

M. Poncino - Politecnico di Torino 33

 Signal Conditioning
• Especially for analog sensors, their output cannot
be used “as is”
• Some conditioning is necessary!
- Amplification
• Signal levels are usually in the µVs to mVs range!
- Filtering
• For noise suppression
and/or frequency selection
(need some specific
band of the signal only)

M. Poncino - Politecnico di Torino 34

 Converting into digital domain
• Regardless of the sensor type the conversion into the digital
domain is mandatory
- Processing IS digital!!!
• A/D conversion implements sampling + quantization
• Key parameter is the
resolution, i.e., # of bits of
the quantization
• Various circuit
implementations!
- We will slightly elaborate
in the corresponding section
Source: Analog Devices

M. Poncino - Politecnico di Torino 35

 The last ring of the chain: Serial interfaces
• Sensors (i.e., ADCs) typically outputs bits serially
(one at a time):
Digital
Sensor
Proc

- Fewer pins
- Better signal robustness (no interference)
- Faster speed at iso-noise
• In the digital word most interfaces are serial!
• Most popular: USB, SPI, I2C

M. Poncino - Politecnico di Torino 39

 Serial interfaces
• USB
- Vast majority of modern devices (less popular for sensors)
- A (host/master) B (peripheral/slave)
- Often used as power source
• I2C (Inter-Integrated Circuit)
- Allows several "slave" (sensor) devices to be accessed by same "master“
- Data (SDA) and clock (SCL) lines (2 wires)
• SPI (Serial Peripheral Interface)
- Multiple slaves, bidirectional
- 4 wires (Clock, select, MISO/MOSI)

M. Poncino - Politecnico di Torino 40

 The overall picture

Computing
device

[Source: P. Tripathi - slideshare] Serial bits

M. Poncino - Politecnico di Torino 41
 Sensor Nodes (aka Smart Sensors)
• How can a non-EE engineer truly use sensors?
- Too much technicalities, interfaces, etc.
• Practical sensors for system designers include all the
components and hide all the details to the designer
- also preventing possible optimizations… SENSOR (SMART)

Digital
serial
output
• Smartness often include
- Programmability (signals to set some parameters)
- Storage (e.g. for config parameters or buffering)
M. Poncino - Politecnico di Torino 42
 Smart Sensors
• Example:
Alphasense
PM sensor

M. Poncino - Politecnico di Torino 43

 Coordinating multiple sensors
• Devices seldom have only one type of sensor
• Multiple sensors can be used to sense independent quantities for
- uncorrelated usage/multiple functions
• E.g., a camera and microphone serve different purposes...
- correlated usage
• To improve the quality of the sensing process
- E.g., gyroscope and accelerometer to improve
an estimate of motion
• Called sensor fusion

Sensor fusion: combining sensory data so that the resulting information has less
uncertainty than would be possible when these sources were used individually.
M. Poncino - Politecnico di Torino 44
 Coordinating multiple sensors
• Sensor fusion is a computational issue, not a sensor one
- Part of the “software” part of an IoT device
- But can be also executed on the sensor

• It requires significant computation complexity

- Feature extraction and in particular classification are complex
algorithms!
- Tradeoff between quality of final classification and complexity of the
method

M. Poncino - Politecnico di Torino 52

 A more systematic view

sensors Signal Digital actua

sensors ADC DAC actuat
sensors cond. processing actuat

Sensing subsystem
Processing subsystem
Actuation (communication) subsystem

M. Poncino - Politecnico di Torino 53

The computing subsystem
(digital domain)

54
 The digital domain
• Fundamentally, digital domain = processing !
• i.e., run algorithms on the sensed/stored data

• A key aspect in IoT devices is the characteristic of the

programmable devices (processors) w.r.t. a traditional
computing systems

Micro-controllers (MCUs)
vs Micro-processors (MPUs)
M. Poncino - Politecnico di Torino 55
 MCUs vs MPUs?
• MPUs contain only a CPU (possibly with some cache)
and interfaces
- They require added peripherals to perform tasks.
• MPUs are quite powerful
- Computational power (multiple execution units (multicore))
- Support for special operations
- Accelerators
• MPUs are ‘modular’
- Resources (memory, disk, etc) are external
- Can be added as needed
• Expensive!!!
M. Poncino - Politecnico di Torino 56
 MCUs vs MPUs?
• MCUs are “complete systems” on a single board or chip
- They contain RAM, ROM, and similar peripherals, which allow them to perform
(simple) tasks independently.
- A number of interfaces typical of sensors or other devices
- Often include one or more ADC for analog inputs!
• MCUs have limited functionalities and modest computational power
• They are meant for application-specific systems
- Simple tasks, possibly repeatedly executed
• Cheap!!

M. Poncino - Politecnico di Torino 57

 Some figures
MPU MCU
Cost 100’s $ 1-10 $
Memory (RAM) A few GBs (A few to 100s) MB
Persistent Storage Disk/SSD, TBs Can be absent, otherwise
FLASH or ROM (kB)
Clock GHz KHz/MHz
Cores 2-16 1 (typical)
Parallelism 32-64 4-32
Speed 100-200 GOPS 1-10 MOPS
Power 100s of W 10-100 uA/MHz
(a few mWs @ max f)

M. Poncino - Politecnico di Torino 58

A different business

M. Poncino - Politecnico di Torino 59

 Conceptual block diagram
MPU = CPU + various MCU = CPU and the “support”
“support” chips/devices devices on a single chip

M. Poncino - Politecnico di Torino 60

 Implication on desgin
• Ex: temperature control device using an MPU and MCU
- Consider the bill of material!!!

M. Poncino - Politecnico di Torino 61

 TI’s CC2650
• “High-end” ultra low-power MCU
• M3 core (16 bit) (<48MHz)
• Includes wireless Xceiver
• 6mA Rx/Tx
• 60uA/MHz!
- 2mA @max speed
• 2.4-GHz RF Xceiver
Compatible With BLE
and IEEE 802.15.4

M. Poncino - Politecnico di Torino 62

 NXP’s LPC1102/1104 (32-bit)
• M0 core
• 12MHz-50MHz Clock
• Vsupply = 3.3V
• On current = 2-7mA

M. Poncino - Politecnico di Torino 63

 Microchip’s PIC MCU
• 7KB RAM!
• 16 bit
• On-current: 2-4mA @3V
CCP = Capture, Compare & PWM
CWG= Complementary Waveform Generator
SMT = Signal Measurement Timer
CLC = Configurable Logic Cell
WDT = Watchdog Timer

M. Poncino - Politecnico di Torino 64

 Other issues
• MCU vs. MPU is a clear dichotomy
- You (usually) have clearly in mind what you need
- MCU selection for your application not a difficult task

• But… there are some non-intuitive issues to be considered and

that are often underrated:
- Programming models/support
- Idle power consumption

M. Poncino - Politecnico di Torino 65

 Programming models and support
• Forget about fancy IDE with nice GUIs
and support for virtually infinite
programming languages
- IDEs are however available…
• You need a host computer (yours) to
program these systems…
- Develop as usual on the host
• But you need to interact with low-level details
- Eventually upload to MCU memory
• SW support/IDE/toolchain and programming models are an
important element for the choice of an MCU!
M. Poncino - Politecnico di Torino 66
Example: STM8l151c3

M. Poncino - Politecnico di Torino 67

 IoT Workloads
• Computational resources are seldom fully utilized
- But typical IoT workloads are sensibly different from traditional ‘computing
ones’ of MCUs
• Duty-cycled workloads
- With very low duty cycles

Active energy Idle energy

M. Poncino - Politecnico di Torino 68
 Idle Power
• For MPUs active power matters
- Active = when the MPU computes
• High utilization rate, a lot of computation
• For MCUs idle power is at least as important
- Idle = when the MCU is waiting for events

Will come back

on this later on!!!
M. Poncino - Politecnico di Torino 69
Example: STM32L4R5

M. Poncino - Politecnico di Torino 70

 A more systematic view

sensors Signal Digital actua

sensors ADC DAC actuat
sensors cond. processing actuat

Sensing subsystem
Processing subsystem
Actuation (communication) subsystem

M. Poncino - Politecnico di Torino 74

Actuation sub-system

75
 Actuation
• The symmetric action of sensing
- DAC + transducer
- Transducer : electrical domain is in input…
• As for sensors, a number of possible devices
- Electric motors
- Comb drives
- Piezoelectric actuator
- Servomechanisms
- Pneumatic actuators
- Hydraulic actuators
- Solenoids
- Stepper motor
- Speakers?
M. Poncino - Politecnico di Torino 76
 Actuation
• One difficulty in driving actuators is that “smart” actuators
(dual of smart sensors) are uncommon
• Need to design the appropriate circuitry for the interface
• Example: driving a speaker
- Speaker essentially a resistor (typical 8Ω)
- Power rating e.g. 1W current P=RI2 I = √P/R = 350mA!!!
- Assuming MCU has analog output, output current in the range of tens
of mA!
• Need an amplifier…

M. Poncino - Politecnico di Torino 77

 “Smart” Actuators
• Some do exist…
• Rely on a common, standardized system architecture that
allows seamless plug&play of sensors and some actuators
• Example: MikroBUS Adapter click boards

M. Poncino - Politecnico di Torino 78

 How about wireless interfaces?
• Can be seen as actuators: Antenna as a transducer
• Standardization helps!
- Wireless interfaces do allow…

• Can think of wireless I/Fs as a special case

of ‘smart actuators’
• Many MCUs include radios and wireless I/Fs
- But we need to understand which is the wireless communication
needed by our application

M. Poncino - Politecnico di Torino 79

 Wireless interfaces
• FOMs
- Bandwidth/bitrate
• correlated to data volume
- Power consumption Will come back
- Range on this later on!!!
- Cost
• Other characteristics
- Architecture: Point-to-point vs. broadcast, # of supported nodes

M. Poncino - Politecnico di Torino 80

 How about displays?
• Again, can be seen as actuators:
• Also here standardization helps!
- Displays I/Fs another special case of ‘smart actuators’

• Depending on the display technology, we can alter the image

to improve power/energy consumption
Will come back
on this later on!!!

M. Poncino - Politecnico di Torino 81