High-efficient energy harvesting architecture for self-powered thermal-monitoring wireless sensor node based on a single thermoelectric generator

Abstract

In recent years, research on transducers and system architectures for self-powered devices has gained attention for their direct impact on the Internet of Things in terms of cost, power consumption, and environmental impact. The concept of a wireless sensor node that uses a single thermoelectric generator as a power source and as a temperature gradient sensor in an efficient and controlled manner is investigated. The purpose of the device is to collect temperature gradient data in data centres to enable the application of thermal-aware server load management algorithms. By using a maximum power point tracking algorithm, the operating point of the thermoelectric generator is kept under control while using its power-temperature transfer function to measure the temperature gradient. In this way, a more accurate measurement of the temperature gradient is achieved while harvesting energy with maximum efficiency. The results show the operation of the system through its different phases as well as demonstrate its ability to efficiently harvest energy from a temperature gradient while measuring it. With this system architecture, temperature gradients can be measured with a maximum error of 0.14 ∘C and an efficiency of over 92% for values above 13 ∘C and a single transducer.

Full text

1
Vol.:(0123456789)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports
High‑ecient energy harvesting
architecture for self‑powered
thermal‑monitoring wireless
sensor node based on a single
thermoelectric generator
Albert Álvarez‑Carulla
1*, Albert Saiz‑Vela
2, Manel Puig‑Vidal
1, Jaime López‑Sánchez
1,
Jordi Colomer‑Farrarons
1 & Pere Ll. Miribel‑Català
1
In recent years, research on transducers and system architectures for self‑powered devices has gained
attention for their direct impact on the Internet of Things in terms of cost, power consumption,
and environmental impact. The concept of a wireless sensor node that uses a single thermoelectric
generator as a power source and as a temperature gradient sensor in an ecient and controlled
manner is investigated. The purpose of the device is to collect temperature gradient data in data
centres to enable the application of thermal‑aware server load management algorithms. By using
a maximum power point tracking algorithm, the operating point of the thermoelectric generator is
kept under control while using its power‑temperature transfer function to measure the temperature
gradient. In this way, a more accurate measurement of the temperature gradient is achieved while
harvesting energy with maximum eciency. The results show the operation of the system through
its dierent phases as well as demonstrate its ability to eciently harvest energy from a temperature
gradient while measuring it. With this system architecture, temperature gradients can be measured
with a maximum error of 0.14
◦
C and an eciency of over 92% for values above 13
◦
C and a single
transducer.
Phenomena such as the internet of things (IoT)1 or wireless sensor networks (WSNs)2 contemplate the installation
of a number of sensors ranging from several hundreds to even thousands2. In these networks, the interest lies in
powering the dierent sensor nodes autonomously. e most direct and widely solution is the use of batteries.
But this has two main drawbacks: (1) an increase in the cost of operation and maintenance of the node, and
(2) a signicant environmental impact each time a battery is disposed. is, multiplied by hundreds of sensor
nodes, makes the long-term sustainability of the network questionable. So powering the node locally by energy
harvesting, enabling internet of battery-less things (IoBT)3, is a better alternative. To ensure the energy viability
of these devices, two ways are used in the node: (1) reduce the number of components, and (2) reduce the com-
putation. e rst strategy is clear: reducing the number of components to be used means fewer components to
be powered and, therefore, less consumption4. e second strategy consists of reducing the computation in the
node5. However, reducing computation at the node in order to make its self-powering possible does not solve
the problem, but rather moves it to another location. In this case, is moved it to data centres.
ey see their activity increased due to an ever-increasing volume of data to be stored and an ever-increasing
computational activity to process this huge amount of data which lead to an expected energy needs up to 1287
TWh in 20306. e aim in these centres is to reduce the energy cost of their cooling or, in other words, to reduce
the power usage eectiveness (PUE). PUE is a standard for measuring the eciency of energy use in data centres
dened by e Green Grid7 and is dened as:
OPEN
1Electronics and Biomedical Engineering Department, Universitat de Barcelona (UB), Marti i Franques, 1-11,
08028 Barcelona, Spain. 2Industrial Engineering and Computer Science, Polytechnic School, University of Lleida
(UdL), C. de Jaume II, 69, 25001 Lleida, Spain. *email: albertalvarez@ub.edu
Content courtesy of Springer Nature, terms of use apply. Rights reserved

2
Vol:.(1234567890)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports/
In an ideal scenario, the PUE would be 1 and all the energy used in data centres would be used solely to power
the servers. However, this is not the case and additional energy is required to power systems such as cooling,
monitoring, lighting, etc. For example, Google data centres have a trailing 12-month PUE of 1.11 in 20208,
Microso a PUE of 1.125 in 20159, Amazon a PUE of 1.2 reported in 201410, and a survey in 2019 revealed that
an average PUE for the other companies of 1.6711. With the goal of improving the current PUE in data centres,
this paper presents a wireless sensor node that uses a thermoelectric generator (TEG) to collect the energy given
o in the form of heat by a microcontroller unit (MCU) in a data centre. Simultaneously, using the same TEG, it
monitors the temperature gradient between the MCU and its environment. is allows monitoring of the ambient
temperature in the data center at a ne granularity that enables the implementation of thermal-aware server load
management algorithms12. e deployment of a wireless battery-less sensor network and the implementation
of thermal-aware server load management algorithms are two pathways that enable improving the PUE. e
data are sent wirelessly with LoRaWAN while the power is extracted from the TEG eciently using a maximum
power point tracking (MPPT) algorithm.
e main lines of research focused on the use of TEGs as a clean power source are the development and fabri-
cation of the TEGs themselves, the design of power management units (PMUs) for ecient energy extraction, and
the design and development of systems and architectures that use TEGs to perform continuous monitoring of a
parameter of interest. For example, a new assembly technology is presented that allows to increase the integration
density of TEGs13. Intended for wearable self-powered devices, a TEG fabricated using this technology allows
achieving higher power values, up to 91 μW for a temperature gradient of 5
◦
C, than other comparable TEGs.
On the other hand, a TEG-based energy harvesting system for powering wireless sensor nodes is presented14. e
system consists of two boost converters and is notable for allowing the system to be started with a voltage of up
to 62 mV. All this, while achieving an eciency of between 44.2 and 75.4%. Another interesting work presents a
self-powered sensor node for monitoring a gas turbine15. e device is capable of delivering a regulated voltage
of 2.4 V and a maximum power of 0.92 W for a turbine temperature of 325
◦
C. And a new MPPT algorithm is
presented to eciently extract energy from TEG16 . e presented method stands out for its simplicity by requir-
ing very few additional components, which also allows to reduce the power consumption of its implementation.
With this method, the TEG is able to work at an operating point deviated by no more than 1.87% with respect
to its maximum power point (MPP).
However, there is not much research on the use of TEG as a power source and sensor simultaneously. is
is also stated in Shi etal. where the authors present one of the few existing works in this regard17. A TEG-based
self-powered wireless sensor node is presented in which the TEG is used as a power source and sensor. e node
consists of a TEG connected to an electronic reader consisting of a PMU, an analog-to-digital converter (ADC),
a microcontroller and a radio frequency (RF) transmitter. e output voltage of the TEG is used as an indicator
signal of the temperature gradient to which it is subjected. is signal is captured and digitized by the ADC and
transmitted wirelessly to a host. However, this approach suers from a drawback that the solution presented
in this work solves. is drawback is that the operating point of the TEG is not controlled. is has two conse-
quences. e rst is that the output voltage of the TEG cannot be used as a direct indicator of the temperature
gradient to which it is subjected. It could be used if the TEG is in open circuit, but this is not the case. Although
the authors indicate that the impedance of the ADC is high, it should be noted that the output of the TEG is also
connected to the PMU that is responsible for powering the entire system. e average consumption of the device,
which is not indicated, makes that the TEG is not in an open-circuit situation. In addition, the dynamic consump-
tion causes the operating point of the TEG to uctuate uncontrollably. is causes an error in the measurement
of the temperature gradient, in this case, of 0.5
◦
C. e second consequence is that energy is not extracted from
the TEG eciently. e TEG needs to operate with a load impedance equal to its internal impedance in order to
deliver maximum power. e use of MPPT algorithms is a must when operating with low temperature gradients
or when the complexity of the system (i.e., its power consumption) is notorious. No eciency value is given
concerning energy extraction from the TEG. Finally, the work indicates a current consumption of 17 mA of the
communication module, but does not describe its design and implementation being a key module of the system.
Another very interesting solution to extract energy and measure temperature using a thermoelectric generator
is presented by Wen etal.18. In this approach two TEGs are used following a double-chain conguration. One has
the role to generate energy and the second one as a sensor instead of our proposal where just one thermoelectric
module is used for both roles at once. e authors use the harvested energy to power a calculator by generating
a regulated voltage of 3.3 V to demonstrate the device’s ability to power commercial solutions. Furthermore, for
the measurement of the power extracted from the TEG, which is 2.9 μW at a temperature gradient of 50
◦
C and
a load resistor of 1.8 k

, the load impedance is set manually without any autonomous matching. is means that
in the event of variations in the equivalent input impedance of the system powered by the TEG, energy is not
extracted from the latter eciently which is other advantage of our system. Finally, it is not shown whether the
solution is able to power a device to enable wireless transmissions, which is one of the main desirable features
of such devices. A calculator is powered which does nothing with the measurement performed by the solution.
In other words, the wearable multi-sensing double-chain thermoelectric generator is able to collect energy and
measure temperature, but it is not able to output the measurement, which makes it far from being a viable com-
mercial solution in the short term.
While the presented solution has a great potential in all applications where continuous monitoring of tem-
perature gradients is required and where it is not feasible to make use of batteries or connections to the power
grid, such as applications in structural health monitoring19,20, temperature gradient monitorization in smart
(1)
PUE
=
Total data center power
IT equipment power
Content courtesy of Springer Nature, terms of use apply. Rights reserved

3
Vol.:(0123456789)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports/
buildings for the generation of energy eciency models21,22, or non-invasive measurement of the temperature
of uids in pipelines using TEGs23, this work presents an approach in which a TEG is operated as a power supply
and sensor simultaneously while extracting energy from the TEG with maximum eciency in the framework of
an application for monitoring the temperature gradient between a server microprocessor and the environment.
at is, at its MPP. For this purpose it controls the operating point of the TEG by using an MPPT algorithm.
Also described is the supercapacitor-based, power-aware wireless transmission management algorithm used to
enable wireless transmission of data via LoRa/LoRaWAN.
Results
The thermoelectric as a generator and sensor. A TEG is a solid state device that transduces heat
energy into electrical energy. As result, an open-circuit voltage (OCV) is originated across the ends of the ther-
mocouple, and a current ow is originated from one end to the other when an electrical load is connected. e
OCV, for a given temperature gradient, is expressed by:
where S is the Seebeck coecient and
T
the temperature gradient across the TEG. Meanwhile, when a load
is attached to the TEG terminals, the TEG output voltage, which depends on its internal resistance
RTEG
, is:
where
RL
is the electrical load. Similarly, the TEG output current can be expressed as:
Derived from Eqs. (2)–(4), Fig.1a shows the simplest lumped electrical model of a TEG24. e measured voltage
vs. current polarization curves of a TGM-127-1.0-2.5 TEG are shown in Fig.1b. e temperature gradient range
used has been selected taking into account the American Society of Heating, Refrigerating and Air-Conditioning
Engineers (ASHRAE) ambient temperature guidelines and the operating temperatures of dedicated micropro-
cessors for servers. e minimum temperature gradient is the critical parameter to validate the feasibility of the
solution. ASHRAE recommends ambient temperatures from 18 to 32
◦
C, which is the most restrictive case and
corresponds to the recommendations for type A1 installations25. As an example of operating temperatures of
dedicated microprocessors for servers, we have taken as a reference a 3rd Gen AMD EPYC processor used by
the main companies that own data centres26. Benchmarking temperature data has been used to establish a mini-
mum operating temperature of 45
◦
C27 and the manufacturer’s specications to establish a maximum operating
temperature of 81
◦
C28. ese temperatures set a minimum temperature gradient of 13
◦
C. In this work, we have
characterized the TEG for a temperature gradient range from 0
◦
C to the maximum temperature gradient that
the used test platform is capable, which is 30
◦
C, with a temperature gradient step of 5
◦
C. e TEG has a See-
beck coecient and an output resistance of 36 mV
◦
C
−1
and 4.4

, respectively. For the minimum temperature
gradient of 13
◦
C, the TEG is capable of generating an OCV of 465 mV, a short-circuit current (SSC) of 108 mA,
and a maximum power at the MPP of 12.7 mW. e performance of a TEG as power source can be evaluated
extracting power versus voltage curves for dierent temperature gradients. From these curves, we can extract
characteristics such as the maximum power for a given temperature gradient, or, more importantly, the operat-
ing point in which this maximum power is extracted. We indicate how eciently the TEG is being operated by
means of the parameter
ηTEG
, dened as:
where
PTEG
is the power extracted from the TEG for a given operating point and
PTEGmax
is the maximum power
that can be extracted at MPP. If the operating voltage of the TEG
VTEG
is normalized with respect the correspond-
ing OCV, as shown in Fig.1c, we can see how the maximum power, i.e., a
ηTEG
eciency of 100%, is achieved for
a polarization voltage equal to one half of the OCV for the same temperature gradient. Usually, from polarization
curves, a voltage vs. temperature gradient curve for a xed load current is extracted to evaluate the performance
of a TEG as sensor. When operated as a sensor, we set the operating point of the TEG in order to maximize its
characteristics as a sensor. For example, to maximize its sensitivity when the voltage is used as output signal, we
set the TEG under an open-circuit operating point. However, when using current as the output signal, we set the
TEG under a short-circuit operating point in order to get maximum sensitivity. In these two operating points,
no power is extracted from the TEG, making unfeasible the usage of the TEG as a power source. With the TEG
as a power source, the MPP must be found using MPPT algorithms.
Finally, Fig.1d shows the maximum power at MPP generated by the TEG for each temperature gradient. e
maximum power shows a dependence on the temperature gradient which is expressed by the function
which is used by the back-end module to compute the measured temperature gradient.
(2)
VTEGOCV
=
ST,
(3)
V
TEG =
ST
RTEG
RL
+1
,
(4)
I
TEG =
ST
RTEG
+
RL
(5)
η
TEG =
P
TEG
PTEGmax
,
(6)
P
TEGmax =74.83 µ
W
C2T2+76.37 µ
W
◦C

T
R
2
=0.9998
Content courtesy of Springer Nature, terms of use apply. Rights reserved

4
Vol:.(1234567890)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports/
Maximum power point tracking algorithms and the maximum power to temperature gradient transfer func-
tion. Several MPPT algorithms exist to track the MPP of a TEG for given operating conditions. But all seek
the same goal: to apply to the TEG a load impedance that matches its internal resistance. From Eqs. (3)–(4), the
condition to reach the MPP is:
that is equivalent to apply a load voltage equal to one half of its OCV. us, one of the most common and
well-known MPPT algorithms is the fractional open-circuit voltage (FOCV) MPPT algorithm29. e TEG is
periodically disconnected from the system to set it in open-circuit condition and sample its OCV. en, the
TEG is reconnected to the system and its operating voltage
VTEG
is regulated to one half of its OCV by means
of the applied load impedance. is algorithm presents one major drawback: no power is extracted when the
TEG is disconnected from the system to measure its OCV. us, the maximum eciency
ηTEG
is achieved while
the TEG is connected and regulated to its MPP, but the average eciency
ηTEG
decreases because of the power
losses introduced into the system during the sampling periods incurring less energy extraction. Furthermore, if
the sampling frequency is reduced to increase
ηTEG
, this can lead to an even lower
ηTEG
because the algorithm
losses its capability to track the variation of the MPP due to variations of the temperature gradient during the
long no sampling periods. Another MPPT algorithm is the perturb and observe (P &O)30. In this algorithm, as
(7)
∂
PTEG
∂RL
=∂
∂RL


(S�T)2
R2
TEG
RL
+2RTEG +RL


(8)
∂P
TEG
∂RL
=0←→ RL=R
TEG
a
b
cd
Figure1. TGM-127-1.0-2.5 thermoelectric generator experimental characterization. (a) Simplied lumped
electrical model of a general-purpose thermoelectric generator. (b) ermoelectric generator’s polarization
curves for dierent temperature gradients (
T
). (c) 3D plot of the power outputted by the thermoelectric
generator (
PTEG
), its eciency (
ηTEG
) and the maximum power points versus its polarization voltage (
VTEG
)
normalized respect its open-circuit voltage (
V
TEG
OCV
), and the temperature gradient (
T
). (d) Transfer function
of temperature gradient (
T
) to power at maximum power point (
P
TEG
max
).
Content courtesy of Springer Nature, terms of use apply. Rights reserved

5
Vol.:(0123456789)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports/
its name states, the TEG output voltage (or current, analogously) is slightly perturbed incrementing or decre-
menting its value while the resulting TEG output power is observed. If the TEG output power increases, the next
perturbation of the TEG output voltage (or current) will be in the same direction. Otherwise, if the TEG output
power decreases, the next perturbation will be in the opposite direction. Resulting from the P &O algorithm, the
TEG operating point will reach its MPP and will oscillate around it. is oscillation of the TEG operating point
around MPP leads to a high but not to a theoretically perfect
ηTEG
. is oscillation can be reduced decreasing
the perturbation step at expense of a slower tracking speed. As benet, in the P &O algorithm, there is no need
to disconnect the TEG from the system to track its OCV or SSC. us, a higher
ηTEG
is achieved.
While the P &O algorithm enables the tracking of the MPP without disconnecting the TEG, the algorithm
can also be used for a completely dierent purpose: to measure the temperature gradient across the TEG. A
temperature gradient versus maximum power (
T
−
PTEGmax
) transfer function can be obtained from the polari-
zation curves. As the P &O algorithm tracks the MPP, the temperature gradient across the TEG can be obtained
from the measurement of the TEG output power, which is already monitored by the algorithm itself. us, only
one single TEG is needed to harvest energy from the environment and to simultaneously sense a temperature
gradient; meanwhile, the former is done eciently. e presented solution makes use of the P &O algorithm to
eciently extract energy from the TEG while simultaneously using it as a temperature gradient sensor.
The self‑powered electronic module. Common approach for TEG-based self-powered devices has three
main modules: (1) a PMU, (2) a front-end and (3) back-end modules. e rst one is responsible for the power
extraction from the TEG. Due to the usually low voltage level outputted by TEGs31, the main task of the PMU is
to boost the input voltage to generate a regulated voltage supply able to power the device. Normally present in
these devices, an auxiliary task is to regulate its equivalent input impedance by means of a MPPT algorithm to
maximize
ηTEG
. is module can be supported with an energy storage module that stores the surplus harvested
energy providing a longer autonomy to the device or the capability to attend a punctual high power requirement.
e front-end module is responsible to interface the sensor used for the temperature gradient measurement,
which can be the TEG or an additional sensor. When the temperature gradient is measured using the TEG itself,
one option is to use the disconnection period during the MPPT algorithm execution to measure the OCV. is
implies a lower
ηTEG
due to the power losses introduced into the system during TEG disconnections. e last
module, the back end, is application dependent, but usually it consists of a MCU, responsible for data processing
and the control of the MPPT algorithm execution; and an interface to output the measurement, such as a wireless
transmitter or a display. e block diagram and a picture of the system presented in this work are shown in Fig.2.
e device uses the P &O algorithm and modulates the equivalent input impedance of its PMU to maximize
ηTEG
. While power is being extracted from the TEG, the front-end or power-sense module, placed in the current
path between the TEG and the PMU, senses the power outputted by the former. is allows to simultaneously
measure the temperature gradient across the TEG. e front-end module outputs two voltage signals
VVTEG
and
VITEG
as indicators of the voltage and current levels outputted by the TEG, respectively. ese two signals
are connected to a MCU that samples them with its analog-to-digital converter (ADC) and computes the TEG
output power
PTEG
. With
PTEG
and
VVTEG
, the MCU controls the execution of the P &O algorithm and, using
its digital-to-analog converter (DAC), generates an analog control signal
VDAC
that is connected to the PMU
closing the feedback loop and controlling its equivalent input impedance. In addition, the MCU converts the
measured TEG output power to temperature gradient and sends it to a wireless transceiver to transmit the data
to a gateway or host. A low-power wide-area network (LPWAN) transmission, like in long-range modulation
wide-area network (LoRaWAN), can require a relatively high-current consumption of 17 mA as minimum32. If
the TEG is not able to meet the required power consumption at the moment of the transmission, the latter will be
unfeasible or, directly, the entire system will be shutdown. To address this, an energy storage module is included
with the PMU. e system harvests energy through the TEG and stores it in the energy storage module. Once
enough energy is stored, the PMU rises the signal
VPGOOD
indicating to the back-end module that the wireless
transmission is energetically feasible and can be done. Once the energy level in the storage module drops, the
signal
VPGOOD
goes down until enough energy is harvested again. In energy storage modules based on a super-
capacitor, the polarization voltage
VSCAP
is used as an indicator of the energy stored dened as:
where
VSCAP
is the voltage across the supercapacitor and
CSCAP
its capacitance.
e choice of the supercapacitor capacitance is a crucial aspect in this type of device because it has a direct
impact on the initial time needed to charge the supercapacitor and to be able to perform the rst wireless com-
munication. In turn, capacitance also inuences the voltage drop caused by a decrease in stored energy. In order
to facilitate system start-up and minimize the time required to perform the rst wireless communication, the
capacitance has been sized to be as small as possible to meet the energy requirements of the wireless communica-
tion. Following Eq. (9), for the same amount of energy a smaller capacitance can be obtained if a larger voltage
drop is allowed. On the other hand, a higher capacitance value would allow a lower voltage drop but would also
lead to a longer start-up time.
Back‑end module consumption. Figure3 shows the current consumption waveforms for both join pro-
cedure and data transmission. To join the LoRaWAN, the back end needs a total energy of 148.9 mJ, while it
needs a total energy of 122.4 mJ to transmit data. In terms of current consumption, the back end consumes 42.6
mA average current consumption during transmission, and 8.6 mA during reception. During idle, when the
MCU is in ultra-low-power operation mode (LPM), the average current consumption is 3.7 μA. Using the most
(9)
E
SCAP =
1
2
CSCAP V2
SCAP
,
Content courtesy of Springer Nature, terms of use apply. Rights reserved

6
Vol:.(1234567890)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports/
restrictive case of 148.9 mJ, a supercapacitor greater than 27.4 mF is required to ensure a maximum voltage drop
at the supercapacitor below 3.3V.
Device start‑up and operation. With the supercapacitor fully discharged, when connected to the TEG
with a temperature gradient applied, the PMU starts to extract power from the TEG and three phases are easily
distinguishable. Shown in Fig.4, phase I corresponds to the period in which the TEG has not yet been connected
to the device and no energy is being extracted from it. Phase II starts as soon as the TEG is connected to the
device. is is when the PMU starts extracting power from the TEG through its charge pump rstly—during
tchgp
—and its boost converter—during
tboost
—once a boosted voltage of 1.8V is reached. When
VSCAP
reaches
a voltage of 5V, the MCU turns on applying the P &O algorithm, and joins the LoRaWAN network. Due to the
OTAA, the energy in the supercapacitor decreases and
VPGOOD
goes high. Once joined, phase III starts and the
ab
c
Figure2. e self-powered thermal-monitoring wireless sensor node prototype based on a single
thermoelectric generator. (a)Block diagram of the system architecture. (b) Photograph of device
implementation. (c) Circuit diagram of the implemented prototype.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

7
Vol.:(0123456789)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports/
temperature gradient is sent every time enough energy is stored in the supercapacitor. Both
ttminTX,TTN
and
tminTX
are shown in the Fig.4. e former corresponds to the minimum period of time that must elapse between
transmissions to comply with eings Network’s fair use policy33. e latter corresponds to the actual mini-
mum transmission period and is only conditioned by the availability of energy in the supercapacitor. From the
experiments performed for dierent temperature gradients, we have measured the time required to charge the
supercapacitor, initially discharged, to 5V and enable wireless communication, denoted as
tstart
−
up
, and the time
required to charge the supercapacitor between transmissions, denoted as
tminTX
as it sets the minimum transmis-
sion period.
tstart
−
up
and
tminTX
have been measured and are shown in Fig.5a–b. e solution presents a start-up
time that shows a dependence on the temperature gradient as could be expected. Higher temperature gradients
allow the TEG to increase the delivered power and speed up the charging of the supercapacitor. For a minimum
temperature gradient of 10
◦
C, the supercapacitance takes 625
◦
C to charge, while for a maximum temperature
Figure3. Current consumption waveforms of the back-end module. (a) During microcontroller’s initialization
and LoRaWAN over-the-air activation. (b) During a LoRaWAN transmission.
a
b
Figure4. Transient waveforms during the start-up and steady-state operation of the device for a temperature
gradient of 13
◦
C. (a) e voltage across the supercapacitor (
VSCAP
), and the signal indicator of the availability
of enough energy to perform a transmission (
VPGOOD
). (b) e regulated voltage supply (
VDD
).
Content courtesy of Springer Nature, terms of use apply. Rights reserved

8
Vol:.(1234567890)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports/
gradient of 30
◦
C, the supercapacitance requires 18 s. e critical start-up time for the minimum temperature
gradient of 13
◦
C is 2.5 min, while for temperature gradients below 10
◦
C the system is not able to start. is
operation limit is due to the minimum cold-start input voltage required by the BQ25504 and the low Seebeck
coecient of the TEG. e transmission period shows the same dependence on the temperature gradient. In
this case, for the critical temperature gradient of 13
◦
C, a transmission period of 44s has been measured. On the
other hand, it has been possible to observe that the device can transmit with temperature gradients above 8
◦
C.
In this case, the temperature gradient is less than the temperature gradient required to start the system. is is
because the BQ25504, once started, requires a lower voltage level. An important aspect of the solution is its abil-
ity to follow the temperature gradient to extract energy from the TEG eciently while being able to obtain the
temperature gradient to which the TEG is subjected. is can be seen in Fig.5c. It can be seen how the prototype
is able to track temperature gradient changes by controlling the TEG operating voltage to one half of its OCV. As
a result, eciencies from 91.8 to 98.8% are achieved, getting an eciency of 92.5% for the critical temperature
gradient of 13
◦
C. In addition, it can be seen how the device is able to calculate the temperature gradient and
transmit it wirelessly with absolute error below 0.22
◦
C, as shown in Fig.5d. Lower
ηTEG
eciencies are shown
for lower temperature gradients. is is because while the TEG operating voltages decrease with applied tem-
perature gradients, the step in voltage used in the P &O algorithm is constant causing there to be a larger relative
oscillation around the MPP for lower temperature gradients. e device performs its operation with a power
consumption of 3.256 mW between transmissions.
Summarising the results obtained, the solution presented is capable of operating from a minimum tempera-
ture gradient of 8
◦
C, a minimum gradient of 10
◦
C being necessary during start-up. For the minimum tempera-
ture gradient, the system is able to start up between 18 and 625s. Using the P &O algorithm, the device is able to
extract energy from the TEG with an eciency of between 91.8 and 98.9% while simultaneously measuring the
temperature gradient with a maximum error of 0.22
◦
C. With a power consumption of 3.256 mW, the sensor node
ab
cd
Figure5. Characterization results of the device. (a) Time required by the device to charge the initially
discharged supercapacitor to 5V to enable wireless communication. (b) Time period between transmissions
versus temperature gradient. (c) ermoelectric generator’s operating (
VTEG
) and open-circuit (
V
TEG
OCV
)
voltages for dierent temperature gradients (
T
). (d) Absolute error of the temperature gradient measurement
and eciency power extraction
ηTEG
.
Content courtesy of Springer Nature, terms of use apply. Rights reserved

9
Vol.:(0123456789)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports/
is capable of transmitting data wirelessly via LoRaWAN (868 MHz) with a power of 14 dBm to reach distances
of up to 10 km line-of-sight distance. Table1 shows a summary of the obtained results compared with other
state-of-the-art solutions considered by the authors to be some of the most relevant ones.
Discussion
A novel approach to measure temperature gradients using a single TEG that, in turn, eciently powers the system
to enable the development of self-powered wireless sensor nodes has been presented. e prototype presented to
validate this approach, is intended to measure the temperature gradients in data centres to apply thermal-aware
server load management algorithms. e oscillation around the MPP is intrinsic to the used P &O algorithm
and results in a gradient temperature measurement error of up to 0.22
◦
C for a temperature gradient of 8
◦
C,
and an eciency of 92.5% for a temperature gradient of 13
◦
C. is is because the voltage step of the algorithm
is constant. A dynamic voltage step would help to improve these specications for low temperature gradients.
Even so, the system is able to operate with a maximum error of 0.14
◦
C for a temperature gradient of 25
◦
C, and
an eciency of up to 98.8% for a temperature gradient of 30
◦
C. e node is able to perform a wireless trans-
mission using LoRaWAN for a minimum temperature gradient of 8
◦
C with a minimum transmission period
of 44s for the critical gradient of 13
◦
C. On the other hand, a minimum gradient of 10
◦
C is required to be able
to start the system. All this is achieved with a power consumption of 3.256 mW between transmissions. ere
are aspects to improve that are part of our future research work. While it is not critical for this application, one
aspect to improve is reducing the minimum system start-up voltage by replacing the currently used commercial
integrated circuit (IC) with a custom integrated version that requires a lower voltage. e second one aspect to
work on is the P &O algorithm so that it applies a dynamic voltage step to help improve the error in the meas-
urement of the temperature gradient and the eciencies obtained. With all this, the presented solution enables
the development of self-powered wireless sensor nodes for monitoring temperature gradients by using a single
TEG operated in an energetically ecient way.
Methods
We have implemented a prototype using current available commercial-o-the-shelf (COTS) discrete compo-
nents. en, we have validated its operation under controlled conditions with a commercial TEG. We describe
the materials and methods used as follows.
Device design and fabrication. We have used a TGM-127-1.0-2.5 from Kryotherm. We have imple-
mented the PMU, the energy storage, and the power-sense modules on a 53mm
×
70mm double-sided printed
circuit board (PCB). A NUCLEO-WL55JC evaluation board from STMicroelectronics has been used to imple-
ment the back-end module. e PMU is based on a BQ25504 IC from Texas Instruments. We have disconnected
the BQ25504’s open-circuit voltage sampling input and have connected a controlled voltage to the IC’s voltage
reference input to regulate the TEG’s output voltage to a desired/controlled voltage value. e harvested energy
is stored on a 30mF supercapacitor from KYOCERA AVX. e supercapacitor value has been selected taking
into account the energy required to perform a wireless transmission and the maximum voltage drop allowed
in the supercapacitor during a period of high energy demand. When enough energy is harvested to perform a
wireless transmission, the PMU turns low the
VPGOOD
signal that is connected to the MCU. Finally, the PMU
has a second dc-dc converter, a LD39050 from STMicroelectronics, is used to generate a 1.8V regulated voltage
supply
VDD
. For the power-sense module, we have used an INA333 instrumentation amplier (InAmp) and a
LPV521 operational amplier (OpAmp) from Texas Instruments to measure the output current of the TEG
ITEG
via a shunt resistor of 0.2

and the TEG output voltage
VTEG
, respectively. ey are routed to dierent input
channels of the MCU’s ADC for conversion and subsequent calculation of the TEG output power
PTEG
and
the corresponding temperature gradient
T
. e back-end module is based on a NUCLEO-WL55JC evalua-
tion board from STMicroelectronics. e module has a 32-bit dual-core STM32WL55JC MCU with essential
features and peripherals for the application like an LPM, a 12-bit ADC with multiple input channels, a 12-bit
Table 1. Summary of the obtained results versus other state-of-the-art solutions.
Parameter Xia etal.34 Brunelli etal.35 Wen etal18 Shi etal.17 is work
Operating temperature gradient range n/a 10–40
◦
C 50–n/a
◦
C 23–n/a
◦
C 8–30
◦
C
Absolute error n/a n/a n/a n/a–0.5
◦
C 0.01–0.22
◦
C
Maximum power point tracking algorithm FOCV None None None P &O
Eciency
ηTEG
19–81% n/a n/a n/a 91.8–98.9%
Power consumption 34.6 μW 6.5 μW n/a n/a 3.256 mW
Start-up time range n/a n/a 19.6–369.0 s 37 s 18–625 s
Data transmission period range 65 s 0–110 s None n/a 10–139 s
Communication protocol ZigBee Simplicity None Gazell LoRaWAN
Base frequency transmission 2.4 GHz 2.4 GHz None 2.4 GHz 868 MHz
Transmission power n/a n/a None n/a 14 dBm
Maximum transmission range
<100
m
<100
m None
<100
m
<10
km
Content courtesy of Springer Nature, terms of use apply. Rights reserved

10
Vol:.(1234567890)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports/
DAC to generate the control signal for the input impedance of the PMU, and a RF transmitter with LoRa. Once
the MCU is powered to its minimum operating voltage, it initializes all its peripherals and proceeds to join a
LoRaWAN using the over-the-air activation (OTAA) method. Once it has joined the network, the MCU enters
in LPM. From then on, the MCU wakes up and samples
VVTEG
and
VITEG
to calculate the power outputted by the
TEG and applies the P &O algorithm. e temperature gradient to which the TEG is subjected is also calculated
using Eq. (6) previously obtained from the TEG characterization. Upon completion of the algorithm iteration,
the MCU returns to LPM. In turn, the MCU has an interrupt congured to know when there is enough power
in the supercapacitor to perform a wireless transmission. is is indicated by the
VPGOOD
signal from the PMU.
Thermoelectric generator and device characterization. A linear voltage sweep measurement has
been carried out on the TEG using a B2962A source meter unit (SMU) from Keysight for dierent temperature
gradients. To set the temperature gradients, a custom platform based on two peltier cells facing each other was
used. e temperature gradient range used has been from 0 to 30
◦
C, with a temperature gradient step of 5
◦
C.
e TEG has been also characterized for the critical temperature gradient of 13
◦
C. In order to select the proper
supercapacitor value, the current consumption of the back-end module has been measured using the B2962A
SMU during the two most energy-demand processes: the OTAA and the wireless transmission. To study the
start-up and steady state operation of the prototype, the voltage at the supercapacitor
VCAP
, the regulated voltage
VDD
, and the signal
VPGOOD
are sampled using an InniiVision 3000A oscilloscope from Keysight. For the meas-
urements, the device with its supercapacitor fully discharged is connected to the TEG subjected to a temperature
gradient. From these measurements, the start-up times
tstart
−
up
and the minimum transmission periods
tminTX
for each temperature gradient are extracted. e former corresponds to the time from the connection of the TEG
until connection to the LoRaWAN. e latter corresponds to the time between transmissions during steady-state
operation. To validate the capability of the device to track the temperature gradient and extract power from the
TEG eciently, the TEG operating voltage
VTEG
and its OCV
VTEGOCV
have been sampled for dierent tem-
perature gradients.
VTEGOCV
has been measured using a second TGM-127-1.0-2.5 TEG thermally connected in
parallel and under open-circuit condition. e B2962A SMU has been used to measure the power consumption
of the device. Finally, to evaluate the accuracy of the system, absolute error have been measured using the same
temperature gradient range that for TEG characterization. e eciency
ηTEG
has been also measured along the
temperature gradient range.
Data availability
e datasets generated during and/or analysed during the current study are available from the corresponding
author on reasonable request.
Received: 18 October 2022; Accepted: 17 January 2023
References
1. Dorsemaine, B., Gaulier, J.-P., Wary, J.-P., Kheir, N. & Urien, P. Internet of things: A denition & taxonomy 72–77. https:// doi. org/
10. 1109/ NGMAST. 2015. 71 (IEEE, 2015).
2. Initiative, I.I. Towards a denition of the internet of things (IoT) (2015).
3. Ju, Q. & Zhang, Y. Predictive power management for internet of battery-less things. IEEE Trans. Power Electron. 33, 299–312.
https:// doi. org/ 10. 1109/ TPEL. 2017. 26640 98 (2018).
4. Alvarez-Carulla, A., Montes-Cebrian, Y., Puig-Vidal, M., Colomer-Farrarons, J. & Miribel-Catala, P. L. Ubiquitous self-powered
architecture for fuel cell-based point-of-care applications. IEEE Trans. Ind. Electron. 68, 11447–11457. https:// doi. org/ 10. 1109/
TIE. 2020. 30294 84 (2021).
5. Xia, S., Yao, Z., Li, Y. & Mao, S. Online distributed ooading and computing resource management with energy harvesting for
heterogeneous mec-enabled iot. IEEE Trans. Wirel. Commun. 20, 6743–6757. https:// doi. org/ 10. 1109/ TWC. 2021. 30762 01 (2021).
6. Koot, M. & Wijnhoven, F. Usage impact on data center electricity needs: A system dynamic forecasting model. Appl. Energy 291,
116798. https:// doi. org/ 10. 1016/j. apene rgy. 2021. 116798 (2021).
7. Grid, T.G. Pue: A comprehensive examination of the metric | the green grid.
8. Google. Eciency-datacenters-google.
9. Microso. Microso’s cloud infrastructure—datacenters and network fact sheet (2015).
10. Amazon. Aws sustainability.
11. Institute, U. 2019 data center industry survey (2019).
12. Villebonnet, V. & Costa, G.D. ermal-aware cloud middleware to reduce cooling needs 115–120. https:// doi. org/ 10. 1109/ WET-
ICE. 2014. 45 (IEEE, 2014).
13. Toan, N. V., Tuoi, T. T. K., Hieu, N. V. & Ono, T. ermoelectric generator with a high integration density for portable and wearable
self-powered electronic devices. Energy Convers. Manag. 245, 114571. https:// doi. org/ 10. 1016/j. encon man. 2021. 114571 (2021).
14. Guan, M., Wang, K., Xu, D. & Liao, W.-H. Design and experimental investigation of a low-voltage thermoelectric energy harvesting
system for wireless sensor nodes. Energy Convers. Manag. 138, 30–37. https:// doi. org/ 10. 1016/j. encon man. 2017. 01. 049 (2017).
15. Wu, Y., Zhang, H. & Zuo, L. ermoelectric energy harvesting for the gas turbine sensing and monitoring system. Energy Convers.
Manag. 157, 215–223. https:// doi. org/ 10. 1016/j. encon man. 2017. 12. 009 (2018).
16. Paraskevas, A. & Koutroulis, E. A simple maximum power point tracker for thermoelectric generators. Energy Convers. Manag.
108, 355–365. https:// doi. org/ 10. 1016/j. encon man. 2015. 11. 027 (2016).
17. Shi, Y. et al. A novel self-powered wireless temperature sensor based on thermoelectric generators. Energy Convers. Manag. 80,
110–116. https:// doi. org/ 10. 1016/j. encon man. 2014. 01. 010 (2014).
18. Wen, D. L. et al. Wearable multi-sensing double-chain thermoelectric generator. Microsyst. Nanoeng. 6(1), 1–3. https:// doi. org/ 10.
1038/ s41378- 020- 0179-6 (2020).
19. Reilly, J. & Glisic, B. Identifying time periods of minimal thermal gradient for temperature-driven structural health monitoring.
Sensors 18, 734. https:// doi. org/ 10. 3390/ s1803 0734 (2018).
20. Newell, S. & Goggins, J. Investigation of thermal behaviour of a hybrid precasted concrete oor using embedded sensors. Int. J.
Concr. Struct. Mater. 12, 66. https:// doi. org/ 10. 1186/ s40069- 018- 0287-y (2018).
Content courtesy of Springer Nature, terms of use apply. Rights reserved

11
Vol.:(0123456789)
Scientic Reports | (2023) 13:1637 | https://doi.org/10.1038/s41598-023-28378-6
www.nature.com/scientificreports/
21. Ahmed, K. et al. Radiant panel and air heating performance in large industrial buildings. Build. Simul. 11, 293–303. https:// doi.
org/ 10. 1007/ S12273- 017- 0414-8/ METRI CS (2018).
22. Uribe, O. H., Martin, J. P. S., Garcia-Alegre, M. C., Santos, M. & Guinea, D. Smart building: Decision making architecture for
thermal energy management. Sensors 15, 27543–27568. https:// doi. org/ 10. 3390/ S1511 27543 (2015).
23. Addabbo, T. etal. Development of a non-invasive thermometric system for uids in pipes. In 2017 IEEE International Symposium
on Systems Engineering, ISSE 2017—Proceedings. https:// doi. org/ 10. 1109/ SYSENG. 2017. 80882 60 (2017).
24. Siouane, S., Jovanović, S. & Poure, P. Equivalent electrical circuits of thermoelectric generators under dierent operating condi-
tions. Energies 10, 386. https:// doi. org/ 10. 3390/ en100 30386 (2017).
25. ASHRAE. ermal Guidelines for Data Processing Environments (ASHRAE, 2021)
26. AMD. Amd epyc
TM
ecosystem partners | amd.
27. Larabel, M. Amd epyc 7763 cooling performance benchmarks. https:// openb enchm arking. org/ (2021).
28. AMD. ermal design guide for socket sp3 processors (2017).
29. Montecucco, A. & Knox, A. R. Maximum power point tracking converter based on the open-circuit voltage method for thermo-
electric generators. IEEE Trans. Power Electron. 30, 828–839. https:// doi. org/ 10. 1109/ TPEL. 2014. 23132 94 (2015).
30. Femia, N., Petrone, G., Spagnuolo, G. & Vitelli, M. Optimization of perturb and observe maximum power point tracking method.
IEEE Trans. Power Electron. 20, 963–973. https:// doi. org/ 10. 1109/ TPEL. 2005. 850975 (2005).
31. Soleimani, Z., Zoras, S., Ceranic, B., Cui, Y. & Shahzad, S. A comprehensive review on the output voltage/power of wearable
thermoelectric generators concerning their geometry and thermoelectric materials. Nano Energy 89, 106325. https:// doi. org/ 10.
1016/j. nanoen. 2021. 106325 (2021).
32. Casals, L., Mir, B., Vidal, R. & Gomez, C. Modeling the energy performance of LoRaWAN. Sensors 17, 2364. https:// doi. org/ 10.
3390/ s1710 2364 (2017).
33. Network, T.T. Duty cycle | the things network.
34. Xia, C., Zhang, D., Pedrycz, W., Fan, K. & Guo, Y. Human body heat based thermoelectric harvester with ultra-low input power
management system for wireless sensors powering. Energies 12, 3942. https:// doi. org/ 10. 3390/ en122 03942 (2019).
35. Brunelli, D., Passerone, R., Rizzon, L., Rossi, M. & Sartori, D. Self-powered WSN for distributed data center monitoring. Sensors
16, 57. https:// doi. org/ 10. 3390/ s1601 0057 (2016).
Acknowledgements
This work was supported by the research Grant PID2019-110142RB-C22 funded by MCIN/
AEI/10.13039/501100011033.
Author contributions
Conceptualization, A.Á.-C.; Methodology, A.Á.-C; Soware, A.Á.-C and A.S.-V.; Validation, A.Á.-C.; Formal
analysis, A.Á.-C. and J.L.-S.; Investigation, A.Á.-C.; Resources, P.Ll.M.-C. and J.C.-F.; Data curation, A.Á.-C.;
Writing - Original dra, A.Á.-C.; Writing - Review & editing, A.Á.-C., A.S.-V., M.P.-V., J.L.-S., J.C.-F. and P.Ll.M.-
C.; Visualization, A.Á.-C.; Supervision, P.Ll.M.-C.; Project administration, J.C.-F. and P.Ll.M.-C.; Funding acqui-
sition, J.C.-F. and P.Ll.M.-C. All authors have read and agreed to the published version of the manuscript.
Competing interests
e authors declare no competing interests.
Additional information
Correspondence and requests for materials should be addressed to A.Á.-C.
Reprints and permissions information is available at www.nature.com/reprints.
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional aliations.
Open Access is article is licensed under a Creative Commons Attribution 4.0 International
License, which permits use, sharing, adaptation, distribution and reproduction in any medium or
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the
Creative Commons licence, and indicate if changes were made. e images or other third party material in this
article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the
material. If material is not included in the article’s Creative Commons licence and your intended use is not
permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from
the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.
© e Author(s) 2023
Content courtesy of Springer Nature, terms of use apply. Rights reserved

1.
2.
3.
4.
5.
6.
Terms and Conditions
Springer Nature journal content, brought to you courtesy of Springer Nature Customer Service Center GmbH (“Springer Nature”).
Springer Nature supports a reasonable amount of sharing of research papers by authors, subscribers and authorised users (“Users”), for small-
scale personal, non-commercial use provided that all copyright, trade and service marks and other proprietary notices are maintained. By
accessing, sharing, receiving or otherwise using the Springer Nature journal content you agree to these terms of use (“Terms”). For these
purposes, Springer Nature considers academic use (by researchers and students) to be non-commercial.
These Terms are supplementary and will apply in addition to any applicable website terms and conditions, a relevant site licence or a personal
subscription. These Terms will prevail over any conflict or ambiguity with regards to the relevant terms, a site licence or a personal subscription
(to the extent of the conflict or ambiguity only). For Creative Commons-licensed articles, the terms of the Creative Commons license used will
apply.
We collect and use personal data to provide access to the Springer Nature journal content. We may also use these personal data internally within
ResearchGate and Springer Nature and as agreed share it, in an anonymised way, for purposes of tracking, analysis and reporting. We will not
otherwise disclose your personal data outside the ResearchGate or the Springer Nature group of companies unless we have your permission as
detailed in the Privacy Policy.
While Users may use the Springer Nature journal content for small scale, personal non-commercial use, it is important to note that Users may
not:
use such content for the purpose of providing other users with access on a regular or large scale basis or as a means to circumvent access
control;
use such content where to do so would be considered a criminal or statutory offence in any jurisdiction, or gives rise to civil liability, or is
otherwise unlawful;
falsely or misleadingly imply or suggest endorsement, approval , sponsorship, or association unless explicitly agreed to by Springer Nature in
writing;
use bots or other automated methods to access the content or redirect messages
override any security feature or exclusionary protocol; or
share the content in order to create substitute for Springer Nature products or services or a systematic database of Springer Nature journal
content.
In line with the restriction against commercial use, Springer Nature does not permit the creation of a product or service that creates revenue,
royalties, rent or income from our content or its inclusion as part of a paid for service or for other commercial gain. Springer Nature journal
content cannot be used for inter-library loans and librarians may not upload Springer Nature journal content on a large scale into their, or any
other, institutional repository.
These terms of use are reviewed regularly and may be amended at any time. Springer Nature is not obligated to publish any information or
content on this website and may remove it or features or functionality at our sole discretion, at any time with or without notice. Springer Nature
may revoke this licence to you at any time and remove access to any copies of the Springer Nature journal content which have been saved.
To the fullest extent permitted by law, Springer Nature makes no warranties, representations or guarantees to Users, either express or implied
with respect to the Springer nature journal content and all parties disclaim and waive any implied warranties or warranties imposed by law,
including merchantability or fitness for any particular purpose.
Please note that these rights do not automatically extend to content, data or other material published by Springer Nature that may be licensed
from third parties.
If you would like to use or distribute our Springer Nature journal content to a wider audience or on a regular basis or in any other manner not
expressly permitted by these Terms, please contact Springer Nature at
onlineservice@springernature.com