ENERGY HAREVSTING USING A
THERMOELECTRIC GENERATOR AND GENERIC
RULE-BASED ENERGY MANAGEMENT
by
YU ZHOU
Submitted in partial fulfillment of the requirements
For the degree of Master of Science
Thesis Adviser: Prof. Swarup Bhunia
Department of Electrical Engineering and Computer Science
CASE WESTERN RESERVE UNIVERSITY
January, 2008
Table of Contents
1. Introduction...... 5
2. Exploiting Wasted Heat in Microprocessor...... 11
2.1 Introduction...... 11
2.2 Modeling...... 15
2.3 Analysis...... 23
2.4 Measurement Results and Applications...... 28
2.5 Summary...... 31
3. Generic Energy Management Platform ...... 32
3.1 Motivation...... 32
3.2 Previous work on hybrid fuel cell and battery system...... 32
3.3 Overall concept, Implementation, Case Study and Results ...... 34
3.3.2 Simulation...... 40
3.3.3 Case Study...... 49
4. Conclusion ...... 57
Bibliography ...... 58
1
List of Tables
No Table Caption Recycled power and the temperature of the hottest spot on the substrate after attaching the TEG on top of the substrate. Two types of I TEG (corresponding to two TEG materials with ZT=1 and ZT=2) are considered. Recycled power and the temperature of the heat spreader after attaching the TEG on the heat spreader. Two types of TEG II (corresponding to two TEG materials with ZT=1 and ZT=2) are considered. Measurement results of energy recycled and the temperature of the III CPU and TEG IV Specifications of proposed energy sources in case study V Specifications of proposed energy users in case study Simulation results for the Case I. No power source is depleted, all VI energy sources are used VII Simulation results for the Case II. Only fuel cell is depleted. VIII Simulation results for the Case III. Only lithium battery is depleted Simulation results for the Case IV. Fuel cell and lithium battery are IX depleted Simulation results for Case V. No energy user is operating in time slot X 1 Simulation results for Case VI. No energy user is operating in time XI slot 2 Simulation results for Case VII. No energy user is operating in time XII slot 3
2
List of Figures:
No Figure Caption
Global primary energy consumption, 1971-2030 Source: Energy 1. White Paper 2005 Japan TEG integrated on the die. It is placed between the package and the 2(a). heat sink TEG integrated on the CPU. It is placed between the package and the 2(b). heat sink Operating model for a TEG. The semiconducting P/N legs (connected 3. in series) generate electricity due to thermal gradient Heat conduction and spreading paths from inside the chip to the 4. ambience 5. Equivalent thermal resistance network for Fig. 4 Thermal resistance network of the heat flow from inside the chip to 6. the ambience with TEG attached to the CPU Measurement setup with a Pentium III processor and a commercial 7. TEG Proposed energy management system along with proposed energy 8. sources and energy sinks 9. Goals of our energy management system 10. Block diagram of energy management system 11. Simulation flow of implemented energy management system 12. I-V characteristic of solar cell [27] 13. Measured fuel cell stack efficiency versus output current [25] Rate capability of QL0700I cell. (a) The discharge curves at different C rates are shown from 4.1 to 2.7V, at 0.2C (thick line), 0.5C (thin 14. line), 1C (dashed line) and 2C (dotted line).
15. An example of time slots for which energy sources are active Power consumption comparison results with and without our energy 16. management technique Depletion point of the fuel cell. With our energy management, the 17. fuel cell can be operated for longer time
3 Energy Harvesting Using a Thermoelectric Generator and Generic Rule-Based Energy Management
Abstract
by
YU ZHOU
Harvesting energy from previously unemployed ambient sources can play an
important role in saving energy and reducing the dependency to primary energy sources
(AC power or battery) of an electronic system. In this work, we investigate harvesting
thermo-electric energy from wasted heat in a microprocessor and propose a generic rule-
based framework for energy management. We develop an analytical model to accurately
estimate the recycled energy considering the non-uniformity of temperature distribution
on the die surface. Further, we propose a possible arrangement for using the TEG on a
processor and provide measurement results on the amount of harvested energy. Next, a
rule-based energy management system is proposed for managing the acquisition, mixing, delivery and storage of energy for any collection of electrical energy sources and
electrical appliances, which have different energy generation and consumption parameters. The proposed energy management system is easily scalable, to cater to a variety of applications with different requirements, while improving the energy utilization and operational lifetime of energy sources.
4 1. Introduction
Due to the continued exploitation of natural resources, the conventional sources of
electric energy, consisting of fossil fuels such as petroleum and coal are getting depleted.
The number of countries that are suffering due to the lack of electric energy is increasing
Fig.1: Global primary energy consumption, 1971-2030. Source: Energy White Paper 2005 Japan.
everyday. Global energy consumption has doubled in the past thirty years and is expected
to increase by another 60% by 2030 [Fig.1]. From the report of International Energy
Agency (IEA) and the Organization of Economic Co-operation and Development
5 (OECD), the consumption rose from 5.5 billion toe (tons of oil equivalent) in 1971, to
10.3 billion toe in 2002. By 2030, global energy demand is expected to reach 16.3 billion
toe, 1.6 times that of 2002 [36]. However, a large portion of this huge energy
consumption is dissipated into the air in terms of heat e.g., from power factory, which can
not be efficiently used by human beings. Hence, a technique to collect this huge amount of wasted heat and convert it to electric energy is worth exploring.
Since the sources of traditional energy consisting of fossil fuels such as petroleum and coal are limited, the increasing tendency to consume these energy sources has increased the importance of renewable energy sources. People have been seeking new
alternative energy sources to replace traditional fossil fuels, such as using solar energy,
wind and wave energy, water energy, geothermal energy and nuclear energy [37].
Although researchers have already explored new energy sources, typically, the
utilization efficiency of previously mentioned alternative energy sources is low. For
example, according to Energie-Fakten [23], coal amounted to 23% of the global energy
sources in 2002, using 3.4 billion tonne coal equivalents (tCE), the major part of which
(2.8 billions tCE) produced 7000 billions kWh of electricity. Nowadays, with a world
average efficiency of around 31%, coal-fired power stations are said to compare
favorably with the upper range of any other power generation technology. Around 70%
of energy obtained from burning coal is dissipated as heat, which means most of the coal
energy is wasted. If we can collect this heat energy dissipated into air and convert even
half of collected heat energy into electricity, we can get double volume of the electricity,
6 compared to the current energy production, which can meet people’s demand for energy
in the next few decades. (Global energy consumption is expected to increase by another
60% by 2030 [Fig.1]). There is abundant heat energy in the nature, e.g. the sun, terrestrial
heat, heat from a car engine etc. Hence, the generation of thermoelectric energy is a very
important issue nowadays. This not only helps to get more energy, but also helps protect
the environment.
Energy harvesting is the process by which energy is captured from environment
and stored in a form which can be put to use. Frequently, this term is applied when speaking about small autonomous devices, like those used in sensor networks [38]. A
variety of different methods exist for harvesting energy based on the energy sources in
the environment, such as solar power, ocean tides, piezoelectricity, thermoelectricity, and
physical motion. In urban areas, there is a surprising amount of electromagnetic energy in
the environment as a result of radio and television broadcasting. Traditionally, electrical
power has been generated from fossil fuels in large, centralized plants. Large-scale
ambient energy from sun, wind and tides is widely available but trickier to harvest. The
history of energy harvesting dates back to the windmill and the water-wheel. Humanity
has also searched for ways to store the energy from heat and vibrations for many decades.
Currently, a major driving force behind the search for new energy harvesting technique is
the desire to power sensor networks and mobile devices without batteries, which get
depleted and have to be recharged or replaced.
7 Energy harvesting devices, which convert mechanical energy into electrical
energy, have attracted much interest in both the military and commercial sectors [39].
Some systems convert random motion, such as that of ocean waves, into electricity to be
used by oceanographic monitoring sensors for autonomous operation. Future applications
may include high power-output devices (or arrays of such devices) deployed at remote locations to serve as reliable power stations for large systems. All of these devices must be sufficiently robust to endure long-term exposure to hostile environments and have a broad range of dynamic sensitivity to exploit the entire spectrum of wave motions.
Energy can also be harvested to power small autonomous sensors such as those
developed using MEMS technology. These systems are often very small and require little
power, but their applications are limited by the reliance on battery power. Scavenging
energy from ambient vibrations, heat or light could enable smart sensors to be functional
for a longer time. Typical power densities available from energy harvesting devices are
highly dependent upon the specific application and design of the harvesting generator.
Renewable energy is the energy derived from resources that are regenerative or
cannot be depleted. For this reason, renewable energy sources are fundamentally different
from fossil fuels, and do not produce as many greenhouse gases and other pollutants as
fossil fuel combustion. A major criticism of some renewable sources is their intermittent
nature. But a variety of renewable sources in combination can overcome this problem.
The challenge of variable power supply may be further alleviated by energy storage.
Available storage options include pumped-storage hydro systems, batteries, hydrogen
fuel cells, and thermal mass. Initial investments in such energy storage systems can be
8 high, although the costs can be recovered over the life of the system. Renewable energy
sources are often dismissed as unreliable. More efficient, diverse, dispersed, renewable
energy systems can make major failures impossible. Storage of energy from renewable
energy systems can also contribute to improved reliability. And one more thing that
people might neglect is that the utilization efficiency of the energy is low. A large amount
of the generated energy may be wasted. Hence, there arises the need for an Energy
Management system, which mixes the energy available from a variety of sources, stores
the surplus energy, routes it to a variety of sinks and efficiently controls the acquisition
as well as the delivery to minimize energy wastage.
In this thesis, we have made the following contributions. We have investigated an energy harvesting technique to recycle the wasted heat in a microprocessor into electric
energy. In particular, we have made the following contributions.
z We have considered two scenarios for placement of a Thermo-Electric Generator
(TEG) on a processor for generating thermo-electric energy: 1) the TEG is directly
integrated onto the substrate (for the best thermo-electric conversion efficiency) as shown
in Fig. 2a; and 2) the TEG is integrated on top of the heat spreader as shown in Fig. 2b.
For the first case, it is important to estimate the TEG efficiency as well as the resultant
die thermal profile (affected due to presence of TEG in the heat-dissipation path),
considering the non-uniform temperature distribution across the die surface. We have
developed analytical models to estimate the efficiency of the TEGs for the first case.
z Using the proposed model and an architecture-level thermal simulator (HotSpot
[17]), we have analyzed the TEG efficiency as well as the temperature of the die surface
9 for varying processor workloads. We have considered both configurations for our
analysis.
z We have presented measurement results from experiments performed with a
commercial TEG and a Pentium III processor, in order to obtain a realistic estimate of the harnessed energy and determine the prospective applications of the recycled energy.
Next, for the energy management platform, we have proposed a generic energy
management system with appropriate algorithm and case study. In particular, we have
made the following contributions.
z We have proposed a rule-based method to dynamically perform energy
management for an arbitrary collection of energy sources and energy users. The rule- based energy management system first collects the energy from active energy sources, mixes them, and then delivers to the energy users based on the algorithm and the rules in the rule library.
z We have also presented simulation results from the proposed energy management
system for different conditions to validate the effectiveness of our proposed approach.
The rest of this thesis is organized as follows. In Section 2, we present our investigations on energy harvesting from wasted heat in a microprocessor by using thermoelectric generators, including the overall concept, modeling, analysis and measurement results. In Section 3, we describe the generic rule-based energy
management platform including motivation, previous work, overall concept and case
studies. Finally, we conclude the thesis in Section 4.
10 2. Exploiting Wasted Heat in Microprocessor
2.1 Introduction
While new sources of energy such as solar energy, wind energy and hydropower etc. are being explored, an important alternate energy source that is often overlooked is thermal energy. Whenever, a work is done, small to large amount of thermal energy is dissipated into the ambience, which if converted back to electric energy may serve useful purposes. This part of the thesis will focus on the use of Thermo Electric Generators for converting wasted heat in high-performance integrated circuits such as microprocessor, into electric energy.
In 1821, Thomas Johann Seebeck discovered that a thermal gradient formed between two dissimilar conductors produces a voltage. At the heart of the thermoelectric effect is the fact that a temperature gradient in a conducting material results in heat flow, which results in the diffusion of charge carriers. The flow of charge carriers to the low- temperature region in turn creates a voltage difference. In 1834, Jean Charles Athanase
Peltier discovered that running an electric current through the junction of two dissimilar conductors could, depending on the direction of current flow, act as a heater or coolant of the junction. The heat absorbed or produced is proportional to the current, and the proportionality constant is known as the Peltier coefficient. Today, based on the knowledge of the Seebeck and Peltier effects, thermocouples have been developed as heaters and coolers as well as thermoelectric generators. Ideal thermoelectric materials have a high Seebeck coefficient, high electrical conductivity, and low thermal
11 conductivity. Low thermal conductivity is necessary to maintain a high thermal gradient
at the junction. Standard thermoelectric modules manufactured today consist of P- and N- doped bismuth-telluride semiconductors sandwiched between two metallized ceramic plates. The ceramic plates add rigidity and electrical insulation to the system. The
semiconductors are connected electrically in series and thermally in parallel.
The foundation of thermal to electric energy conversion rests on the Seebeck
effect, which involves the generation of an electromagnetic force (emf) when the two
junctions of two dissimilar metal bars connected to each other are kept at different
temperatures [1]. The structure consisting of dissimilar metals is often referred to as a
thermo-electric generator (TEG). Researchers have already attempted to exploit this
effect for generation of electric energy from thermal energy. In [2], the authors have
developed a semiconductor based TEG with P-N legs for energy conversion. The
measurement results presented in [2] indicate energy conversion efficiency as high as
40%, by applying a temperature difference larger than 100°C. In [3], a TEG has been
developed with an output voltage as high as 6.4V for a temperature level of 250°C-350°C
at the hot side of the TEG. These works demonstrate that thermo-electric conversion
using a TEG can be a promising energy harvesting application.
Modern high-performance chips are operating at multi gigahertz frequency, such
as microprocessors, which consume large amounts of power (in the order of 40-100W)
[20]. The dynamic power is increasing due to increase in operation frequency and the
leakage power is also increasing due to the technology scaling and increasing operation
12 temperature. A substantial part of consumed power is translated into heat. This heat creates a large temperature gradient between the die surface and environment. In order to ensure reliable operation of the die at elevated temperature, we need to design appropriate heat removal mechanism using high-efficiency heat spreader and heat-sink. A relevant question in this context would be: Can we exploit the thermal gradient in a high- performance chip to recycle the wasted heat energy into thermoelectricity using thermo electric generators?
In this part, we have performed the modeling, analysis and measurement of the thermo-electric energy conversion in relation with a modern microprocessor and a commercial TEG. The electric energy continuously recovered from this wasted heat during the operation of the processor can be used to drive other components in a system or effectively stored for future use. Interestingly, the temperature distribution on the die surface is non-uniform (comprising of localized hotspots [17]) leading to a reduced thermo-electric conversion. The concept of using TEGs to generate electric energy from the wasted heat of a microprocessor was first proposed by Suski in a patent [4] and the feasibility was evaluated in [5] – [6]. In this work, TEG was attached directly on the CPU and on the other side of the TEG a heat sink with cooling fan was attached.
Older thermo-electric generation devices typically used bi-metallic junctions, but most thermoelectric devices currently in use generate electricity utilizing semiconductor materials (such as Bismuth Telluride, Bi2Te3), which are good conductors of electricity but poor conductors of heat [12]. These semiconductors are typically heavily doped to
13 create an excess of electrons (n-type) or a deficiency of electrons (p-type). An n-type semiconductor will develop a negative charge on the cold side and a p-type semiconductor will develop a positive charge on the cold side, which forms a current flowing from one semiconductor leg to another. Since each P-N leg of a semiconductor thermoelectric device will produce only a few millivolts, it is useful to connect these legs in series to generate higher electric voltage. Researchers are also attempting to manufacture TEGs with high thermal conductivity, so that an integrated TEG can be an effective replacement for the heat spreader. However, due to limitations in the nature of the materials used for building the TEGs, the efficiency of the present-day TEG is typically less than 10% [13].
However, this recycled electric energy can be stored in a super-capacitor and reused later or can be used to drive low-power portable electronics such as MP3 players or PDA, which only consume about 110mW and 200mW, respectively [22]. Although the current efficiency of the TEG is low (less than 10%), with the advancement of technology, we can get high efficiency TEGs and use them in a computer to harvest wasted energy from the microprocessor and use it to drive other components. Hence, it is necessary to develop an accurate model for the TEG and the die thermal profile, which can predict its efficiency and detect whether the junction temperature is below a threshold after attaching the TEG.
14
Fig. 2a: TEG integrated on the die. It is placed between the package and the heat sink.
Heat Sink Heat TEG Flow TEG in Shunt Heat Spreader Die PCB Fig. 2b: TEG integrated on the CPU. It is placed between the package and the heat sink.
2.2 Modeling
Previous work presented in [6, 7] have tried to analyze the efficiency of a TEG
which is directly attached to the CPU (Fig. 2a) or in a shunt setup (dashed box in Fig. 2b).
In both cases, the surface of the TEG in contact with the CPU has been considered to be
at a constant temperature. However, in reality, due to localized hotspots the die
temperature is non-uniform and calculations based on the constant temperature profile
will lead to inaccurate prediction of the TEG generated voltage.
15 2.2.1 TEG Efficiency Considering Non-uniform Temperature Distribution on Die
Surface
At the steady state, the heat generated from the CPU is equal to the heat dissipated, so that the temperature of the CPU remains constant and the amount of heat received as input by the TEG can be considered to be a constant value. At this steady state condition, it is possible to model the TEG efficiency by determining the amount of heat that is converted into electric energy. Fig. 3 shows an operational model for the TEG, where the open circuit voltage is given by Equation (1).
UNo =Δiiα T (1)
In Equation (1) Uo is the open circuit voltage, N is the number of P/N leg pairs, α is Seebeck coefficient, ΔT is the temperature difference between two sides of the TEG.
The power generated by the TEG is given by Equation (2).
Fig. 3: Operating model for a TEG. The semiconducting P/N legs (connected in series) generate electricity due to thermal gradient.
16 2 22⎛⎞NTiiα Δ 2 ⎛⎞⎛⎞UNTo iiα Δ PLL==IRiii⎜⎟⎜⎟ R L = R L =⎜⎟2iiiNLρ i R L (2) ⎝⎠⎝⎠RRLPN++ RR LPN ⎜⎟RL + ⎝⎠A
In Equation (2), RL is the load resistance, ρ is the density of the materials used to
manufacture P/N legs, L is the length of one P/N leg and A is the cross-section area of
one P/N leg. Under the condition of output load matching, the maximum power delivered
by the TEG is given by Equation (3).
2 2 ⎛⎞NTiiα Δ ()NTiiα Δ PRL = ⎜⎟i PN = (3) ⎝⎠RRPN+ PN 4RPN
With the above three basic equations, it is now possible to take into account the non-
uniform temperature distribution at the hot side of the TEG in contact with the CPU. We
first partition the floorplan of the processor into number of different functional units such
as integer unit, floating unit, cache, etc. If there are ‘m’ partitions in total and unit m has
nm P/N leg pairs in contact with it, then the total open circuit voltage generated by the
TEG can be represented as:
Un11=Δiiα TUn 1,22 =Δ iiαα T 2,, … Unmm = ii Δ T m
The total voltage UTotal is:
UUUUUtotal=++++123... n
=Δ+Δ++ΔnTnT1122iiα iiαα...... nmm ii T
=Δ+Δ++Δα(nTnT1122ii ..... nmm i T ) (4)
The total power generated by considering the non-uniform die temperature
distribution can therefore be calculated as:
17 2 U total P L = 4 R PN
2 2 α (nTn1122iiΔ+ Δ T + ..... + nmm i Δ T ) = 4 R PN
22 α (nTn1122iiΔ+ Δ T + ..... + nmm i Δ T ) = ρ i L (5) 8 (nn123+++ n ...... + nm ) A
ρiL where, RnnnnPNm=++++2 (123 ...... ) (6) A
Thus, given the number of P-N leg pairs per unit area and the area for each partition on the die floorplan, it is possible to accurately calculate the power generated
from the TEG considering the non-uniform temperature distribution. A more practical
configuration involving the generation of electricity from the wasted heat of the processor
is shown in Fig. 2b, where the TEG is attached to the heat spreader layer. We will
compare the effectiveness of energy recovery between the two configurations in Section
2.3.
2.2.2 Die Thermal Profile with Integrated TEGs
In both cases, it is important to calculate the die temperature profile in presence of
the TEG. The primary reason is that due to the low thermal conductivity of the TEG, the
thermal resistance in the heat dissipation path increases, which results in less amount of
heat being dissipated to the environment in unit time, leading to an increase in the die
18 conv R Rsink TEG R spreader R
Fig. 4: Heat conduction and spreading paths from inside the chip to the ambience.
temperature. The steady-state thermal profile of the die will depend on the TEG material
and heat load from the processor.
The previous work presented in [8] – [11] which model the heat conduction and
spreading within the package constitute the basis for our estimation of the die thermal
profile in presence of the TEG. Inside the package of the chip, a three-dimensional heat flow exists from the device layer to the ambience. Fig. 4 shows the heat dissipation path
from inside the chip to the ambience, and Fig. 5 shows the equivalent thermal resistance
network along which the heat dissipates. As seen from Fig. 4, the heat suffers refraction
when it moves from a layer with thermal conductivity k1 to k2, the angle of refraction θ is
given by Equation (7) [8].
−1 k1 θ =tan ( ) (7) k 2
Since each layer inside the package behaves as a heat source for the layer above it,
we calculate the thermal resistance of the TEG using the formula for thermal resistance
19 presented in [9], which takes into account the two-dimensional spreading of heat. The
resistance, as given by Equation (8) considers x and y to be the length and width of the heat source, and L and k to be thickness and the thermal conductivity of the layer in
contact with the heat source.
12tanyL+ θ x R =⋅⋅ln 2kxyxLy tanθθ (−+ ) 2 tan (8)
The total thermal resistance along the heat dissipation path as shown in Fig. 4 and
5 is the summation of the thermal resistance of each component. For example, when the
TEG is attached to the heat spreader, the total thermal resistance is represented by
Equation 9, where Rsubstrate, Rspreader, RTEG, Rsink and Rconv are the thermal resistances of
substrate, heat spreader, TEG, heat sink and convection, respectively.
RR=++++substrate R spreader R TEG R sink R conv. (9)
Rsubstrate, Rspreader, and RTEG can be calculated by using equations 7 and 8. For
Rsubstrate, the area of the heat source is equal to the area of each functional block and the spreading angle can be determined by the thermal conductivity ratio of the heat spreader
and silicon substrate. The thermal resistance of the heat spreader and TEG can be
estimated using the same method. However, the area of the heat source for heat spreader
and TEG is the original heat source area plus an area expansion due to heat spreading as
shown in Fig. 4. For heat spreader, the length and width of the heat source is:
(2xL+ substrate tan)θ substrate (2yL+ substrate tan)θ substrate
For TEG, the length and width of the heat source is:
20 (xL++ 2substrate tanθ substrate 2 L spreader tanθ spreader )
(yL++ 2substrate tanθ substrate 2 L spreader tanθ spreader )
The spreading angle is related to the thermal conductivity of the heat spreader and
substrate [8].
Due to the spreading effect in the intermediate layers, every section of the heat
sink receives the same amount of heat and therefore it suffices to calculate the total
resistance of the heat sink, instead of sections on top of each functional unit. The thermal
resistance of the heat convection is given by Equation (10), where h is the heat
convection coefficient and A is the effective area of the heat sink. The method for
calculating effective area of the heat sink with straight fins is described in [12]. 1 Rconv. = hA⋅ (10)
From Fig. 4, we find that the heat dissipation paths merge after a point. This phenomenon indicates that the steady-state temperature of the layers, which are along the heat dissipation path but are away from the die surface, will depend on the power consumption of all on-chip heat sources as given by Equation 11.
N Tx(,y )=⋅∑ RQii (11) i=1
In Equation 11, T(x, y) is the temperature at location (x, y) on the surface of an
intermediate layer, which may be either heat spreader or the TEG. Ri is the thermal
21
(a) (b) Fig. 5: Equivalent thermal resistance network for Fig. 4.
Fig. 6: Thermal resistance network of the heat flow from inside the chip to the ambience with TEG attached to the CPU. resistance between heat source i and location (x, y). Qi is the power consumption of heat source i, N is the total number of on-chip heat sources. From Fig. 4, we can also derive an equivalent thermal resistance network as shown in Fig. 5. In Fig. 5, Q1 and Q2 are the heat generated by two heat sources. T1 and T2 are the temperature of two heat sources.
The heat dissipation paths for these two heat sources are initially separated but will merge finally due to heat spreading. Due to heat spreading, temperature of the surface of an intermediate layer will be uniform, which is indicated by the equivalent parallel resistor
Re in Fig. 5b. The distance at which heat from different sources merge together can be estimated based on the heat spreading angle in each packaging layer and the distance between each heat source. The junction temperature T1 and T2 can be estimated as:
TQRQQR11112=++()e (12) TQRQQR22212=++()e
22 Hence, the general term of the equation for calculating junction temperature by
considering inter-heat source correlation can be estimated as:
TQRQQQiii=+++++(....)123 QR ne (13)
Ren= RRR122// // //...... // R
th where Ti is the junction temperature of i heat source, Qi and Ri are the heat generated by
the ith heat source and the thermal resistance along the heat dissipation path for the ith heat
source before the heat from different sources merge together. Thus, based on Equation
(8), it is possible to calculate the thermal resistance of each intermediate layer. Equation
(13) provides the temperature of the heat sources on the die after the TEG is attached to the heat spreader. The equivalent thermal resistance network corresponding to the heat flow from inside the chip to the ambient is shown in Fig. 6.
2.3 Analysis
In this section, we will analyze two possible configurations discussed in Section II
in terms of their effectiveness of energy conversion. One is to attach the TEG to substrate
(Fig. 2a) and the other is to attach the TEG to the heat spreader (Fig. 2b). Before we analyze the configurations, it is necessary to define a figure of merit for the TEG, a higher value of which translates to a higher TEG efficiency. Such a merit (referred as
“ZT”) as defined in Equation (14) indicates that good thermoelectric materials should have large Seebeck coefficient α, higher electrical conductivity σ, higher hot side
temperature Th, and low thermal conductivity λ and can therefore achieve higher
efficiency for thermo-electric conversion.
23 α 2σ ZTT= h (14) λ
2.3.1 TEGs Attached to Substrate
In this subsection, we have analyzed the effect of attaching the TEG directly to
the silicon substrate of the chip (Fig. 2a). Since the dimensions of the TEG are taken to
be same as that of the die, it has fewer P-N leg pairs compared to the case where the TEG
is attached to the heat spreader. According to the thermal resistance models developed in
the previous section, we have calculated the temperature distribution of the substrate after
attaching the TEG. Power trace files for an Alpha 21264 microprocessor were obtained
by simulating different SPEC95 benchmarks on Wattch architecture level performance
and power simulator (version 1.0) [21]. Using the power trace information for each
functional block of the processor and Equations 7-13, it is possible to estimate the
junction temperature of each functional block. For calculating the power and the
temperature, a few assumptions were made about the dimensions of the die and the TEG.
Because power trace is obtained from an Alpha 21264 microprocessor, we use the
dimensions of this microprocessor for simulation [17]. The dimension of the die (as well
as of the TEG) is taken to be 18mm*18mm with a 0.5mm thickness. The dimension of the
heat sink is assumed to be 60mm*60mm with a 6.9mm thickness. The height and width of
the P-N leg is assumed to be 1mm and 0.5mm, respectively. With this dimensional data, it
is now possible to calculate the number of P-N legs that are in contact with each functional block. Other parameters like Seebeck coefficient (α), thermal conductivity of
TEG (λ) and its electrical resistivity were obtained from [6], where the reported value for
ZT is 0.9. Calculations have also been made on the basis of the parameters presented in
24 [13], where a value of ZT=2 has been reported. Table I reports the power generated by
the TEG and the highest junction temperature on the die after attaching the TEG to the
substrate for ZT=1 and ZT=2, respectively.
For ZT=1 in Table I, we see that the highest junction temperature decreases after
the TEG is attached to the substrate. Such a fall in temperature can be attributed to the
fact that the thermal conductivity of the TEG is very low, about 100 times lower than that
of the substrate. As we had mentioned previously, heat undergoes refraction when it
moves from one medium to another. The refraction angle is related to the thermal
conductivities of the two mediums. Due to the large difference in the thermal
conductivity of these two media, the heat refraction angle is very large, so that the heat generated by each functional block of the microprocessor will spread to the entire TEG, which suggests a uniform temperature distribution on the TEG. Due to high thermal
resistance of TEG, the non-uniform temperature distribution of the die is also alleviated
because the large amount of heat generated by the localized hotspot can be transferred to
the cooler region on the die through the TEG. In spite of this heat transfer, the
temperature of the cooler region remains almost constant due to its large area compared
to the hotspots. For the APPLU benchmark, the temperature difference between the
hotspot and the cool region is only about 10°C before the TEG is attached to the substrate,
which is almost same as the decrease in the temperature of the hottest spot on the
substrate.
25 For a TEG with ZT = 2, the thermal conductivity of the TEG material is much lower compared to a material corresponding to ZT = 1. In this case, due to the very high thermal resistance of the TEG, the amount of heat that can be transferred through the
TEG is significantly reduced. Thus the temperature of the hottest region of the substrate substantially increases (Table I) after the TEG is attached to the substrate.
2.3.2 TEGs Attached to Heat Spreader
In this scenario, the TEG is attached on the heat spreader, which has a larger area than the die. The dimension of the TEG is taken to be same as that of the heat spreader, i.e. 30mm*30mm with a thickness of 1mm. The dimensions of the die and heat sink are same as in the previous case. Calculations were performed for both ZT=1 and ZT=2.
Other parameters of the TEG such as Seebeck coefficient, thermal conductivity, electrical resistivity etc. were kept unchanged. The output power and the temperature of the heat spreader are provided in Table II.
From Table II, we can see that the power generated by the TEG decreases on attaching the TEG to the heat spreader. In this scenario, the TEG is attached to the heat spreader, so the temperature at the hot side of the TEG decreases compared to the previous scenario, and if the ambient temperature is kept constant, the temperature gradient across the TEG reduces. Although the number of P/N legs increases due to an increase in the area of the TEG, temperature gradient across the TEG reduces significantly. This cannot be compensated by the increase in the number of P-N legs.
Compared to the heat spreader, the thermal resistance of the TEG is quite large, which
26 translates into an additional large thermal resistance between the heat spreader and heat
sink. This reduces the heat transfer rate from the heat spreader to the ambience, resulting
in an increase in temperature of the heat spreader (Table II).
Table I. Recycled power and the temperature of the hottest spot on the substrate after attaching the TEG on top of the substrate. Two types of TEG (corresponding to two TEG materials with ZT=1 and ZT=2) are considered.
SPEC-95 Recycled power (mW) Initial Final temperature of the hotspot Benchmark ZT=1 ZT=2 temperature ZT=1 ZT=2 APPLU 132 369 135.12°C 127.37°C 168.84°C APSI 55 153 101.85°C 96.20°C 122.82°C CC1 77 214 114.68°C 106.71°C 138.22°C Compress95 80 224 95.58°C 92.26°C 117.04°C Go 123 344 138.92°C 125.18°C 165.23°C hydro2d 111 311 126.98°C 120.11°C 158.15°C Li 102 285 123.95°C 116.74°C 153.14°C M88ksim 126 352 133.25°C 125.40°C 165.88°C Perl 109 304 127.41°C 119.81°C 157.43°C turb3d 124 345 132.20°C 124.49°C 164.58°C wave5 106 295 124.90°C 118.17°C 155.12°C
Table II. Recycled power and the temperature of the heat spreader after attaching the TEG on the heat spreader. Two types of TEG (corresponding to two TEG materials with ZT=1 and ZT=2) are considered.
SPEC-95 Recycled power (mW) Initial Final temperature of the spreader Benchmark ZT=1 ZT=2 temperature ZT=1 ZT=2 APPLU 26 76 44.33°C 64.53°C 76.25°C APSI 11 31 42.67°C 55.74°C 63.27°C CC1 15 44 42.77°C 58.64°C 67.55°C Compress95 9 27 42.36°C 54.65°C 61.67°C Go 24 71 43.99°C 63.68°C 75.01°C hydro2d 22 64 43.97°C 62.50°C 73.25°C Li 20 59 43.87°C 61.52°C 71.82°C M88ksim 25 73 44.01°C 63.94°C 75.39°C Perl 21 63 43.92°C 62.25°C 72.88°C turb3d 24 71 44.02°C 63.71°C 75.04°C wave5 21 61 43.91°C 62.15°C 72.39°C
27
2.4 Measurement Results and Applications
2.4.1 Measurement Results
In order to find out the amount of energy harnessed from the wasted heat of a microprocessor, experiments were carried out to determine the range of power generated by a commercial TEG in a practical scenario. The experimental setup consisted of an
Intel Pentium III processor running Windows XP system applications (no user applications) at 1GHz with the heat sink and cooling fan removed. A thin copper plate
(used as heat spreader) was attached on the package of the CPU by using a thermal gel in order to make a good thermal contact between the CPU and the copper plate. A Bi-Te based commercial TEG [18] was then attached on the copper plate. Fig. 7 shows our experimental setup. The CPU lying beneath the copper plate is depicted by a red bulge on the Cu plate. The TEG rests on the other end of the Cu-plate, which forms a shunt for the heat to be transferred from the CPU to the TEG. The shunt method (which is shown as a dashed box in Fig 2b can provide an additional parallel heat dissipation path, compared to directly attaching the TEG above the CPU.
The experiments were then carried out for four different scenarios which are as follows: 1) the Cu plate rests on the CPU and the TEG rests on the Cu shunt away from the CPU, both Cu plate and the TEG being exposed to the ambience. 2) The position of the TEG is same as before, except that the surface of the TEG not in contact with the Cu plate is kept in contact with a cooler surface. This allows a higher temperature gradient across the TEG, allowing higher energy conversion compared to scenario 1. 3) The Cu-
28 Processor (underneath the Cu heat spreader)
TEG
Fig. 7: Measurement setup with a Pentium III processor and a commercial TEG.
Table III. Measurement results of energy recycled and the temperature of the CPU and TEG.
Test condition Temp. of Temp. of Temp. of Voltage Current Impedance matched CPU Cu plate TEG (mV) (mA) power (mW) TEG Scenario I 77°C 43°C 40°C 87.7 14.5 0.3 on shunt Scenario II 77°C 43°C 37°C 200.1 30.1 1.5 TEG Scenario III 77°C 59°C 53°C 210.3 31.6 1.7 on Scenario IV 77°C 59°C 47°C 418.8 64.3 6.7 CPU
plate still rests on the CPU and TEG is attached on the section of the Cu-plate which is
exactly above the CPU, and the TEG is exposed to the ambience. 4) The configuration is
similar to scenario 3, except that the upper surface of the TEG is cooled using a cold
surface to increase the thermal gradient.
Table III presents the power and temperature values obtained from our
measurements for the four cases discussed above. In order to verify the correctness of our
measurement, we have also calculated the expected values of the generated power based
29 on the specifications of the commercial TEG. On an average, we find the measured
power values are lower by about 20% compared to the expected ones. The discrepancy between the measured and the expected values can be attributed to the fact that we have only measured the temperature difference between the bottom layer and the top ceramic layer of the TEG, which is not indicative of the actual temperature difference that exists between the two metal junctions. In reality, the temperature difference will be lower than the measured value. Moreover, the estimated power value assumes a perfect thermal contact between Cu plate and the TEG, which is difficult to achieve in the experimental setup. The amount of recycled power depends on 1) the TEG efficiency and 2) temperature of the cooler side of the TEG. Scenario II and IV support that if the open side of TEG is cooled, more power can be recycled. Note that, the maximum conversion efficiency is determined by the Carnot efficiency, which is about 4% assuming a temperature difference of 12°C and a high temperature of 59°C.
2.4.2 Application to an Electro-Osmosis System
Although the recycled energy is only several mW (as shown in Table III), it can
be increased considerably using high-efficiency TEG materials that possess higher
electrical conductivity but lower thermal conductivity and cooling system in the open
side of the TEG. Recently, different thermoelectric modules based on novel materials and
structures (such as superlattice systems) and their potential applications have been
reported [14] – [15]. In [15], the authors have developed a TEG with a maximum
thermoelectric conversion efficiency of 5.6%, which was applied to collect the wasted
heat from a bulb in a projector system and operate cooling fans and other electronic
devices.
30
A possible way to reuse the harvested thermoelectricity is to drive an electro- osmosis system to cool the CPU. Electro-osmosis, which involves the motion of a polar liquid through a membrane under the influence of an applied electric field [16], is being considered as an efficient cooling mechanism for modern microprocessors [19]. For an electro-osmosis system, based on dimensions, the driving voltage may vary from mV to several volts. Based on specification of the microprocessor system, it is, therefore, possible to design an electro-osmosis system that operates at low voltage and power supplied by the TEG. The system would essentially work in a feedback loop, where an increase in the die temperature would lead to higher TEG output, which can potentially increase the liquid flow (positive feedback) thereby improving the cooling capacity.
2.5 Summary
Harvesting wasted heat energy from a microprocessor system can be effective to increase the efficiency of the energy usage for a computer system. In this section, we have presented a model to accurately estimate the TEG efficiency by considering the non-uniform temperature distribution on the die surface. Models to estimate the final temperature of the die surface after attaching the TEG to the substrate are also presented.
Using our model and existing architecture-level power/thermal simulators, we have analyzed the TEG efficiency and die temperature for different processor workloads.
Finally, experiments were carried out to measure the power that can be generated by a commercial TEG in a realistic scenario, and suggest potential applications for such thermo-electric systems. Emerging TEGs with large Seebeck coefficient and higher
31 thermal resistance as well as better cooling at the cooler side of the TEG can help to
increase the amount of recycled energy significantly.
3. Generic Energy Management Platform
3.1 Motivation
Due to the increasing power requirement, power consumption of electronic equipment has emerged as a major problem in electronic system design. To minimize the power consumption, improve the energy utilization efficiency and extend the lifetime of
the energy sources, an energy management system can be developed to effectively
manage the energy between sources and users. This energy management system should
be generic which can be used for wide range of sources and appliances, varying in size
and power generation/consumption ranges. For example, it can be used for managing
power delivery in mobile or wearable electronic devices (e.g. cellular phone, MP3
players, PDAs etc) as also for mobile sensor networks. It can also be used for managing
power supplies for household appliances like washing machines, TV, stereo system,
refrigerator etc. Other applications include power management for electronic devices in a
car or a navigation system or in an industrial environment.
3.2 Previous work on hybrid fuel cell and battery system
Researchers have already developed some techniques for power management to
reduce power consumption and improve energy utilization efficiency. Dynamic power
management (DPM) is an effective and well-known technique [25] to reduce energy
consumption at the system level. It puts the device into a low-power state when the
32 system is idle or requires lower energy, so as to minimize the total energy consumption of the system. A number of previous works have targeted prediction of future idle periods
[28, 29, 30], stochastic control [31, 32], and aggregation of small idle times to get longer idle durations [33, 34]. While the power management strategies in existing works target energy minimization of the whole system, typically they do not take into account the characteristics of the energy sources. As a result, the minimum energy consumption of the system may not necessarily transform to the maximum lifetime of the overall system.
Notable exceptions are the battery-aware power management strategies that explicitly take into account the battery non-linearities by battery scheduling [32] and load profile shaping [35]. While conventional power management techniques minimize the energy consumption of the embedded system, they do not consider the properties of the energy sources. Alternative energy sources such as fuel cells (FCs) have substantially different power and efficiency characteristics that have to be taken into account when developing strategies that maximize their power consumption and operational lifetime. A fuel- efficient DPM policy is described in [25], which aims at maximizing the operational lifetime of the FC by jointly applying DPM on the embedded system and fuel-efficient current setting of the energy source. Maximizing the lifetime of the FC is equivalent to minimizing the fuel consumption in a given period of time. They determine FC output setting by utilizing an optimization framework that considers the FC system efficiency characteristics explicitly. For run-time operation, they propose the fuel-efficient DPM algorithm, FC-DPM, which applies the optimal FC output setting policy.
33 3.3 Overall concept, Implementation, Case Study and Results
Fig. 8 shows the overall flow of our proposed energy management system. In the
proposed system, we consider three energy sources — Solar Cell, Fuel Cell and Lithium
Battery and three energy users — Laptop, GPS (Global Positioning System) and PSP
(Play Station Portable). The energy management system should perform the following functions:
i) Acquire Energy
The system will collect energy from all kinds of energy sources, e.g. solar panel,
rechargeable battery, fuel cell, etc. Keeping in mind that different sources have different
power delivery features e.g. different voltage and current levels, different ranges of
efficiency and even different times of operation, the system should have the capability to
mix the different forms of energy with different parameters.
ii) Deliver Energy
The system will dynamically “route” energy from appropriate energy sources to
energy users, based on their capabilities and requirements. It may need to connect
multiple sources to a particular sink or redirect the energy from a single source to
multiple sinks. Based on the requirements of the different energy users it is supposed to
serve, the energy management system will automatically decide how to route the energy.
34 iii) Store Energy
The system will store the excess energy in a suitable storage device (e.g. super-
capacitors) for future use.
Apart from these basic functions, the system should be able to make control
decisions to maximize energy saving at the sinks or minimize the energy loss at the
acquisition end or routing paths. The system should also be responsible for the
maintenance of the energy sources and sinks to ensure longer lifetimes, by dynamically
turning off energy sources when not required and by cutting off power supply to energy
sinks, when not in use, to eliminate power dissipation due to standby leakage current.
Energy Sources Energy Sinks
Solar Cell Notebook
CPU Redirect Fuel Cell GPS
Storage
PSP Battery
Fig. 8: Proposed energy management system along with proposed energy sources and energy sinks.
35 It should also control the individual energy sources like the fuel cell or solar
panels based on the energy requirement and their energy generation capability. For
instance, the energy delivery from a fuel cell can be controlled by varying the rates of
O2/H2 discharge, depending on the energy requirement at that point of time. Again,
energy sources (e.g. solar, vibration and thermal) may not be functional all the time.
Different batteries also have different characteristics like varying discharge rate between
alkaline and rechargeable lithium-ion batteries, where the latter can maintain a constant
but small current level for a sustained period of time. The energy must be harvested
efficiently from these sources and stored for future use. Novel techniques for energy harvesting from heat energy (dissipated in a microprocessor or other electronic chips using TEG) or from kinetic energy involved in motion or from vibration energy of heartbeats are being developed and can be incorporated as new sources of energy.
Similarly, the energy consumers also have different energy requirements at
Fig. 9: Goals of our energy management system. different times. For instance, a microprocessor may require a sudden burst of current
36 depending on the activity going on at that time. This requirement may not satisfied by a
source, which can only provide low current levels. Buffers might be inserted in the power
delivery path or the energy management module can mix energy from multiple sources to
satisfy the system requirements at that point of time. Hence, it should be capable of
providing different energy levels to different sources at different times. In case of
overload (when energy available is less than energy required), the control system must take priority decisions so that some essential appliances continue to receive energy, instead of shutting off power supply to all users simultaneously.
Another desired characteristic of the energy management system is scalability. It should be able to accommodate large number of sources and sinks, with variable spatial and temporal requirements. This will be explained in more details when we describe the scope of applications of this system.
Fig. 10: Block diagram of energy management system.
37
The goal of our proposed energy management is to achieve better efficiency, compared to conventional energy delivery systems. We consider four aspects to characterize the efficiency of this system: 1) increasing the operational lifetime of energy sources and sinks; 2) minimizing the energy loss in the routing path; 3) maximizing the availability of the energy sources; and 4) saving energy. Fig. 9 illustrates the major goals of our proposed system.
3.3.1 Implementation
The system consists of three major components: a CPU, a Redirect unit and a
Storage unit (as shown in Fig. 10). These three components along with some sensors constitute the hardware platform of this energy management system. The sensor 1 detects the activity of the energy sources. If there is any available energy source, the sensor will send a signal to the CPU to let it know the current energy generation capability of the energy source (e.g. the intensity of light for solar cell) and/or the amount of capacity still left in the energy source (e.g. amount of fuel left in the fuel cell). The sensor 2 will detect the activity of energy users. If there is any active energy user, the sensor 2 will send a signal to the CPU to let it know how many energy users are active, the workload of each energy user and the required current to drive them. The sensor 3 detects the activity of the storage cell. If the storage cell, e.g. super capacitor is not full, the sensor will send a signal to the CPU. The excess energy from the energy sources that is larger than the power requirement of active energy users will be delivered to the storage unit to charge it. If the storage is full, the sensor will send a signal to the CPU to stop storing. If the active energy sources fail to satisfy the total power requirement of
38 active energy users, the energy stored in the storage unit will be delivered to them. The
CPU acts as the brain of the system and makes all decisions based on the signals from all
the sensors and the pre-defined rules in the rules library. The decision will be sent to the
energy delivery network to turn on/off some switches making sure that the energy from
the energy sources is delivered properly to the energy users. The power delivery network
is basically a switch circuit network. It receives the control signals from the CPU and
turns on/off its switches to deliver the power to the energy users or storage unit. Between
the power delivery network and energy users, there is a DC-DC converter. This is
because the specifications (described in details in the following section) in terms of
current and voltage output/input of the energy sources and energy users may not match.
The voltage requirement of energy sources are sometimes larger than those of the energy users, hence a DC-DC converter might be required to convert the high voltage to the low voltage that the energy users can use. Also, when choosing the proper DC-DC converter,
Fig. 11: Simulation flow of implemented energy management system.
39 Table IV. Specifications of proposed energy sources in case study Voltage Peak Current Capacity Source Name Max Current (mA) (V) (mA) (mA.h) Solar Cell 12 1000 1200 Infinite Fuel Cell 18.2 1200 1500 Finite Lithium Battery 10.8 2200 3000 Finite
we should ensure that the output current can meet the current requirement of all energy
users.
3.3.2 Simulation
We have written a C program to simulate the functional behavior of this rule-
based energy management system. The work flow is illustrated in Fig. 11. The processing
unit will take the information from three files—Energy Source, Energy User and
Operation (i.e. workload), then based on the existing Rules, the Processing Unit will make the decision, which includes the information about connections between energy
sources and energy users at certain time points, the total power consumed by energy users
and the capacity left in the energy sources. The following part will introduce the
specifications of the energy sources, energy users and the various operations considered
for this study.
40
3.3.2.1 Energy Sources
The energy source file contains the information on voltage, peak as well as maximum output current and the capacity. In our case study, we take three energy sources: solar cell, fuel cell and lithium battery. Table IV shows the specification of these energy sources. We take the specification of solar cell from [24] and consider the
ENCAPSULATED type Solar Cell (2V/200mA). We connect six solar cells in series to obtain these specifications. The specification of fuel cell is taken from [25] and the specification of lithium battery is from an operational battery used by a laptop. The
format of the energy source file is as follows:
Format:
User Name: XXXX
Voltage: XXXX
Current Voltage Plot for PV Module 1000 900
800 700 600 IV curve 500 Power (Watts*10) 400 300 Current (milliamps) 200 100 0 0 5 10 15 20 25 30 35 Voltage (Volts)
Fig. 12: I-V characteristic of solar cell [27].
41 Peak Current: XXXX
Maximum Current: XXXX
Capacity: XXXX
3.3.2.1.1 Solar Cell
Solar power is a renewable energy source, which can have infinite capacity depending on environmental state. It is already widely used in home and business [26].
Fig. 12 shows the I-V curve for a typical solar cell. From the curve we can observe that, the solar cell is similar to a battery and as the load varies, the current-voltage curve does not follow Ohms law. The maximum power is delivered at a voltage of 25V. Hence, to maximize its efficiency, we also choose the peak operating point and maximum point of the solar cell. In the solar cell we have considered, the peak current is 1000mA and maximum current is 1200mA.
3.3.2.1.2 Fuel Cell
Fig. 13: Measured fuel cell stack efficiency versus output current [25].
42 From [25], we know that a fuel cell package can generate power longer (4 to 10X) than a battery package of the same size and weight. But, the power and efficiency characteristics of the FC are quite different from batteries. The variation in efficiency is much larger for fuel cells. From Fig. 13, we can observe that the efficiency is decreasing for curve (a). When the output current is larger than 1200mA, the efficiency decreases drastically; so we choose 1200mA as the peak current for the fuel cell and 1500mA as its maximum current.
3.3.2.1.3 Lithium battery
For a lithium-ion battery, the discharge rate impacts the total capacity. Fig. 14 is a
curve of voltage versus discharge capacity. Based on this curve and the specification of the lithium battery, we choose about half of the total capacity as the peak current which is
2200mA and 2500mA as maximum current.
Fig. 14: Rate capability of QL0700I cell. (a) The discharge curves at different C rates are shown from 4.1 to 2.7V, at 0.2C (thick line), 0.5C (thin line), 1C (dashed line) and 2C (dotted line).
43
3.3.2.2 Energy Users
The energy users we choose are laptop, GPS and PSP. The energy user file contains the information about voltage and current requirements at different workloads.
Table V contains the specifications of energy users. The format of energy user file is as follows:
Format:
User Name: XXXX
Voltage: XXXX
Current at high workload: XXXX
Current at medium workload: XXXX
Current at low workload: XXXX
3.3.2.3 Operations
The operation file contains the information on activity (start or end), operation Table V. Specifications of proposed energy users in case study
User Name Voltage (V) High Current (mA) Medium Current (mA) Low Current (mA)
Laptop 10.8 2000 1000 200
GPS 5 60 30 5
PSP 3.6 800 200 10 time point and the workload of every operation. The format of operation file is as follows:
User Name: XXXX
44 Activity: XXXX
Time Point: XXXX
Workload of the operation: XXXX
3.3.2.4 Rules
The Rule file provides the instructions to the CPU for making the decisions. It contains the priority list of all energy sources (in our simulation, the priority order is solar
cell → fuel cell → lithium battery) and the decisions corresponding to different conditions. The main algorithm of the energy management followed by the CPU is described as follows:
*************************************************************
*****
INPUTS: powersource_specifications, poweruser_specifications, activity list for each
powersource and user
OUTPUT: Energy consumption up to each time instant, Power delivered at each time
instant,
Current provided by each powersource and consumed by each poweruser
documented in an output file along with error messages and/or warnings
45 *************************************************************
*****
function CYCLE( ) /*cycle-accurate simulator*/ begin
clk:=0;
n=0;
while(true)
begin
clk:=clk+1;
if (clk=time of nth activity OR any energy source is turned on or off)
begin
i_total := total current requirement for all the active users at that instant;
call POWER_MGMT( );
n := n+1;
end if
if (clk=MAX_CLK)
begin
break from while loop;
end if
update values of remaining capacity of each energy source having limited capacity and total energy consumed till that time instant;
46 if (remaining capacity of any energy source is close to zero) /*If it does not have enough charge to drive current for the next second*/
begin
turn off that energy source;
end if
end while end
The algorithm of the function--POWER_MGMT( ) is described as follows:
*************************************************************
*****
INPUTS: i_total, i_provide /*existing values for each energy source*/
OUTPUT: i_provide /*updated values*/
*************************************************************
*****
function POWER_MGMT ( ) /*Power management function based on simple adaptive
rules*/
begin
for each energy source in order of priority
begin
if (energy source is on)
begin
if (energy source's peak operating current >= i_total)
47 begin
i_provide(for that source) := i_total;
i_total := 0;
break from for loop;
else
begin
i_provide(for that source) := i_peak(for that source);
i_total := i_total - i_peak(for that source);
end if
end if
end for
if (i_total > 0)
for each energy source
begin
if (energy source is on)
begin if (energy source's maximum supply current - energy source's peak operating current
>= i_total)
begin
i_provide(for that source) := i_total;
i_total := 0;
break from for loop;
else
48
Fig. 15: An example of time slots for which energy sources are active.
begin
i_provide(for that source) := i_max(for that source);
i_total := i_total - (i_max(for that source) - i_peak(for that source));
end if
end if
end for
end if end
3.3.3 Case Study
We selected a set of cases to simulate the behavior of our proposed rule-based energy management system. The simulation time window is one day i.e. 24 hours. The time slots for active energy sources are shown in Fig. 15.
• From 00:00:00 to 05:59:59, only lithium battery is available.
• From 06:00:00 to 17:59:59, solar cell, fuel cell and lithium battery are all
available
49 • From 18:00:00 to 23:59:59, fuel cell and lithium battery are available
3.3.3.1 Case I (no energy source is depleted, all energy sources are used)
In this case, all energy sources are used by the energy users, but the capacity of the fuel cell and lithium battery is sufficient so as not to be depleted by the energy users.
In this case, we assume the capacity of the fuel cell is 50000mA.h and lithium battery is
40000mA.h, respectively. The operation specification for this case is as follows:
Laptop start 3 hour 34 minute 56 second high
Laptop end 8 hour 34 minute 56 second high
Laptop start 9 hour 34 minute 56 second medium
Laptop end 11 hour 34 minute 56 second medium
GPS start 2 hour 34 minute 56 second high
GPS end 12 hour 54 minute 56 second high
GPS start 17 hour 23 minute 45 second medium
GPS end 20 hour 23 minute 23 second medium
PSP start 6 hour 34 minute 56 second low
PSP end 9 hour 34 minute 57 second low
PSP start 12 hour 34 minute 45 second high
PSP end 15 hour 23 minute 12 second high
50 Power Consumption Comparision (Case I)
51000 49000 47000 45000 43000 41000 39000 Capacity left (mA.h) 37000 35000
6 0 4 7 0 5 :0 3 :56 :5 4:56 0:0 34 :34:56 :23:12 :23:23 2:34: 3:34:56 6:00 6:34: 8:34:56 9:34 9: 2:5 11 12:34:451 15 17:23:4518:0 20 Time Cap._left_FC (mA.h) Cap._eft in FC without EM (mA.h)
Fig. 16: Power consumption comparison results with and without our energy management technique.
The simulation results are shown in Table VI (in the appendix). During this operation, with the proposed energy management system, the fuel cell consumption can be reduced by 71.80%. And the energy saved will vary depending on the operations.
Fig.16 compares the power consumption between the case with energy management and the one without energy management. From this figure, we can observe that there is more charge left in the fuel cell with our energy management technique. And depending on different operations, the energy saved varies unless the fuel cell is depleted. The power consumption curves in Cases V, VI and VII are quite similar to the curve in this case, which depicts the basic trend.
51 3.3.3.2 Case II (only fuel cell is depleted)
In this case, the fuel cell is depleted. We modified the energy source file and set a
low capacity (10000mA.h) for the fuel cell. The operation file is still the same as in Case
I. The simulation results are presented in Table VII in the appendix. From this table, we
can see that at 7:52:28, the fuel cell is depleted. But without our energy management, the
fuel cell is depleted much earlier than 7:52:28, so the life time of the fuel cell can be
extended. In this case, the fuel cell is depleted, so we compare the operation time for the
fuel cell. Fig. 17 shows that the fuel cell can be used for a longer time with energy
management.
3.3.3.3 Case III (only lithium battery is depleted)
Power Consumption Comparision (Case II)
2500
2000
1500
1000 Depletion point without energy management 500 Depletion point with energy management 0 Capacity lef (mA.h) 6 6 5 5 -500 4: 4: 4:56 3:45 0:01 :3 :3 :3 :23:12 :2 :00:00 :0 02 03:34:5606:00:0006:34:5607:52:2808:34:5609 09:34:5711 12:34:4512:54:5615 17 18 18 20:23:23 -1000 Time
Cap._left_FC (mA.h) Cap._eft in FC without EM (mA.h)
Fig. 17: Depletion point of the fuel cell. With our energy management, the fuel cell can be operated for longer time.
52 In this case, the lithium battery is depleted. We modified the energy source file and set a low capacity for the lithium battery, which is 2000mA.h and we also set the capacity of fuel cell to 50000mA.h. The operation file is unchanged. The simulation results are presented in Table VIII in the appendix. From this table, we can observe that the lithium battery is depleted at 4:31:25. If there are no other active sources, the energy users cannot be supplied with power until other active energy sources become available after 6:00:00.
3.3.3.4 Case IV (lithium battery and fuel cell are depleted)
In this case, the lithium battery is depleted. We modified the energy source file and set a low capacity for the lithium battery, which is 2000mA.h and we also set the capacity of fuel cell to 3000mA.h. The operation file is not changed from the one in Case
I. The simulation results are presented in Table IX in the appendix. Fuel cell and lithium battery are both depleted. From 07:52:28, the solar cell will output its maximum current
1200mA to the energy users. We can also note that, with our energy management, the fuel cell can operate for a longer duration.
3.3.3.5 Case V (no energy user is operating in time slot 1)
In this case, we change the operation to ensure that there is no energy user active in time slot one. We set the capacity of the fuel cell and battery to 50000mA.h and
40000mA.h, respectively. The operation specification for this case is as follows:
53 Laptop start 7 hour 34 minute 56 second high
Laptop end 8 hour 34 minute 56 second high
Laptop start 9 hour 34 minute 56 second medium
Laptop end 11 hour 34 minute 56 second medium
GPS start 10 hour 34 minute 56 second high
GPS end 12 hour 54 minute 56 second high
GPS start 17 hour 23 minute 45 second medium
GPS end 20 hour 23 minute 23 second medium
PSP start 6 hour 34 minutes 56 second low
PSP end 9 hour 34 minute 57 second low
PSP start 12 hour 34 minute 45 second high
PSP end 15 hour 23 minute 12 second high
The underlined entries indicate the modifications we made, compared to the operation in
Case I. The simulation results are presented in Table X in the appendix. In this case, the
power saved by fuel cell is 82.66%. From Table X we note that the longer the fuel cell
operates, the more energy it can save.
3.3.3.6 Case VI (no user is operated in time slot 2)
In this case, we change the operation to ensure that there is no energy user active
in time slot two. We keep the capacity of the fuel cell and battery the same as in the previous case. The operation specification for this case is as follows:
54 Laptop start 7 hour 34 minute 56 second high
Laptop end 8 hour 34 minute 56 second high
Laptop start 20 hour 34 minute 56 second medium
Laptop end 22 hour 34 minute 56 second medium
GPS start 4 hour 34 minute 56 second high
GPS end 5 hour 54 minute 56 second high
GPS start 19 hour 23 minute 45 second medium
GPS end 20 hour 23 minute 23 second medium
PSP start 19 hour 34 minute 45 second high
PSP end 20 hour 23 minute 12 second high
The simulation results are presented in Table XI. We can see the power reduction is only
38.64%, which is because the total operation time in this case is very short.
3.3.3.7 Case VII (no user is operated in time slot 3)
In this case, we change the operation specifications to ensure that there is no
active energy user in time slot three, keeping the capacity of the fuel cell and battery as in the previous two cases. The operation specification for this case is as follows:
Laptop start 7 hour 34 minute 56 second high
Laptop end 8 hour 34 minute 56 second high
Laptop start 9 hour 34 minute 56 second medium
55 Laptop end 11 hour 34 minute 56 second medium
GPS start 10 hour 34 minute 56 second high
GPS end 12 hour 54 minute 56 second high
GPS start 15 hour 23 minute 45 second medium
GPS end 27 hour 23 minute 23 second medium
PSP start 6 hour 34 minute 56 second low
PSP end 9 hour 34 minute 57 second low
PSP start 12 hour 34 minute 45 second high
PSP end 15 hour 23 minute 12 second high
The simulation results are presented in Table XII. The power saving is 63.34% due to longer operation time compared to case VI.
Based on all the simulation results, we can validate that the behavior of our proposed rule-based energy management system matches the expected behavior and we also achieve savings in terms of the power and lifetime of operation for the fuel cell.
Further, this algorithm is generic - thus, we can add new energy sources and/or energy users into the system simply entering their specifications in the appropriate files and updating the rule library.
56 4. Conclusion
We have investigated energy harvesting from wasted heat in a microprocessor and propose a generic rule-based framework of energy management. In the energy harvesting part, first, we develop an analytical model to accurately estimate the recycled energy considering the non-uniformity of temperature distribution on the die surface. Next, we analyze the effectiveness of the approach for thermo-electric generator (TEG) with different efficiencies (measured in terms of its figure of merit, ZT) under varying processor workload. Finally, we propose a possible arrangement for using the TEG on a processor and provide measurement results on the amount of harvested energy. The measurements on a Pentium III processor running at 1GHz show that we can harvest
~7mW of power from the processor for average workload using a commercial TEG.
Emerging TEG devices with higher ZT can increase the efficiency of recycling the wasted heat significantly. A possible application of the harvested energy would be to drive a low-power electro-osmosis system to cool the processor. In the second part, we propose a generic rule-based energy management system for managing the acquisition, mixing, delivery and storage of energy for an arbitrary collection of electrical energy sources and electrical appliances, which have variable parameters of energy generation and consumption. In the rule-based energy management system, we have proposed a simple rule-based approach to perform energy management dynamically. The system gathers energy from active energy sources; mixes it, and then delivers it to the energy users based on their current status (obtained from built-in sensors) and the set of rules in the rule library. We also performed several case studies to simulate the behavior of the
57 energy management system under different operating conditions. Simulation results
validate the effectiveness of the proposed approach.
The initial system of energy management developed here can be significantly
enhanced. First, in our program, the rule library is hard-coded. To make it more flexible,
the rules should be input from a file. Second, we have only considered some simple rules
(such as priority of one energy source over another and the peak efficiency point of the
energy sources). More advanced rules can be incorporated into the rule library to improve
the energy utilization efficiency. The system can also be augmented to “learn” based on
the energy usage pattern and modify its rule database dynamically. When a new condition
occurs, for which no rule exists in the rule library, the system can create a new rule and
adaptively absorb this rule into the library to use it next time. In the program, we did not
consider energy storage unit, which can be charged to store the excess energy and
discharged when no active source is available, due to time limit. Finally, more
simulations need to be performed to further validate the correctness of decision-making
and power delivery capability of the system under different conditions. We also plan to
build a micro-controller based hardware prototype system using discrete components for
validating the energy-management scheme.
Bibliography
[1] http://en.wikipedia.org/wiki/Seebeck_effect#Seebeck_effect
[2] Kucherov, Y, et al., “Energy Conversion Using Diodelike Structures,” International Conf. on
Thermoelectronics, 2002.
58 [3] Anatychuk, L.I., “Thermoelectric Generator Modules and Blocks,” International Conference on
Thermoelectronics, 1997.
[4] Suski, Edward D., “Method and Apparatus for Recovering Power from Semiconductor Circuit Using
Thermoelectric Device,” US Patent #5,419,780, 1995.
[5] Yazawa, K. et al., “Thermoelectric-Powered Convective Cooling of Microprocessors,” IEEE Trans.
on Advanced Packaging, 2005.
[6] Solbrekken, G.L. et al., “Thermal management of portable electronic equipment using thermoelectric
energy conversion,” ITHERM, 2004.
[7] Angrist, S. W., “Direct Energy Conversion”, 3rd ed., Allyn and Bacon, Inc., Boston, MA, 1976.
[8] Nguyen, N. B., “Properly implementing thermal spreading will cut cost while improving device
reliability,” In Proc. Int. Symp. Microelectronics, 1996.
[9] Meeks, D., “Fundamentals of heat transfer in a multilayer system,” Microwave Journal, Jan. 1992.
[10] Song, S. et al., “Closed-form equation for thermal constriction/spreading resistances with variable
resistance boundary condition,” Electronics Packaging Conference, 1994.
[11] Incorpera, Frank P. and Dewitt, David P., “Fundamentals of Heat and Mass Transfer,” 4th Edition,
John Wiley & Sons, 1996.
[12] Bass, John C., “Thermoelectric Generator,” United States Patent 5892656, 1999.
[13] Sethumadhavan, Simha et al., “Powering a Cat Warmer Using Bi2Te3 Thin-Film Thermoelectric
Conversion of Microprocessor Waste Heat,” ASPLOS, 2006.
[14] Chu, Rencai et al., “Thermoelectric Generator utilizing Boiling-Condensation (Experiment and
Modeling)”, Proc. 22th Int. Conf. on Thermoelectrics, 2003.
[15] Horio, Yuma et al., “Performance and Application of Thermoelectric Modules for Consumer Use
Fabricated with (Bi,Sb)2(Te,Se)3 using a Rapid Solidification Technique,” International Conference
on Thermoelectrics, 2005.
[16] http://en.wikipedia.org/wiki/Electro-osmosis
[17] http://lava.cs.virginia.edu/HotSpot/
[18] http://www.thermonamic.com/Pspec.html
[19] http://www.patentstorm.us/patents/6942018-description.html
59 [20] Mahajan, Ravi et al., “Cooling a Microprocessor Chip,” Proceeding of IEEE, 2006
[21] http://www.eecs.harvard.edu/%7Edbrooks/wattch-form.html
[22] González, J.L et al., “Human Powered Piezoelectric Batteries to Supply Power to Wearable
Electronics Devices,” IJSMER 2002
[23] http://www.euractiv.com/en/energy/analysis-efficiency-coal-fired-power-stations-evolution-
prospects/article-154672
[24] http://www.electronics123.com
[25] Zhuo, Jianli et al., “Dynamic Power Management with Hybrid Power Sources”, Design Automation
Conference, 2007
[26] http://www.sunpowercorp.com/Default.aspx
[27] http://solardat.uoregon.edu/download/Lessons/PVLessonPlan2SolarElectricArrays.doc
[28] C.-H. Hwang and A. Wu, “A predictive system shutdown method for energy saving of event-driven
computation,” in Proc. of ICCAD, Nov. 1997, pp. 28–32.
[29] M. Srivastava, A. Chandrakasan, and R. Brodersen, “Predictive system shutdown and other
architectural techniques for energy efficient programmable computation,” IEEE Transactions on
VLSI Systems, vol. 4, pp. 42–55, 1996.
[30] EY Chung, L. Benini, and G. De Micheli, “Dynamic power management using adaptive learning
tree,” in Proc. of ICCAD, Nov. 1999, pp. 274–279.
[31] L. Benini, A. Bogliolo, and G. De Micheli, “A survey of design techniques for system-level dynamic
power management,” IEEE Transactions on VLSI Systems, vol. 8, pp. 299–316, 2000.
[32] P. Rong and M. Pedram, “Battery-aware power management based on Markovian decision
processes,” IEEE TCAD, vol. 25, pp. 1337–1349, 2006.
[33] R. Jejurikar and R. Gupta, “Leakage aware dynamic voltage scaling for real-time embedded
systems,” in Proc. of DAC, 2004, pp. 275–280.
[34] Y. Lu, L. Benini, and G. De Micheli, “Low-power task scheduling for multiple devices,” in Proc. of
CODES, 2000, pp. 39–43.
[35] L. Benini, G. Castelli, A. Macii, and R. Scarsi, “Battery-driven dynamic power management,” IEEE
Design and Test of Computers, vol. 18, pp. 53–60, 2001.
60 [36] Japan Agency for Natural Resources and Energy. Energy White Paper 2005 (Tokyo, 2005).
[37] http://www.altenews.com/Alternative%20Energy%20Overview.pdf
[38] Jason Hu,“ Energy Harvesting Support for Sensor Network”, IPSN 2005
[39] http://intel-alliance.org/signal/articles/anmviewer.asp?a=1279&print=ye
61
APPENDIX
62
Table VI. Simulation results for the Case I. No power source is depleted, all energy sources are used.
Cap._left in FC I(Lap) I(GPS) I(PSP) I(tot) I(SC) I(FC) I(Battery) Ch(tot) Ch(FC) Ch(Battery) Cap._left_FC Cap._left_Battery Time Ch(SC) ( C ) without EM (mA) (mA) (mA) (mA) (mA) (mA) (mA) ( C ) ( C ) ( C ) (mA.h) (mA.h) (mA.h) 02:34:56 0 60 0 60 0 0 60 0.00 0.00 0.00 0.00 50000.00 40000.00 50000
03:34:56 2000 60 0 2060 0 0 2060 216.00 0.00 0.00 216.00 50000.00 39940.00 50000
06:00:00 2000 60 0 2060 1000 1060 0 18146.24 0.00 0.00 18146.24 50000.00 34959.38 50000
06:34:34 2000 60 10 2070 1000 1070 0 22464.00 2096.00 2221.76 18146.24 49382.84 34959.38 49126.67 63 08:34:56 0 60 10 70 70 0 0 37368.00 9296.00 9925.76 18146.24 47242.84 34959.38 46126.67
09:34:56 1000 60 10 1070 1000 70 0 37620.00 9548.00 9925.76 18146.24 47242.84 34959.38 46126.67
09:34:57 1000 60 0 1060 1000 60 0 37621.07 9549.00 9925.83 18146.24 47242.82 34959.38 46126.25
11:34:56 0 60 0 60 60 0 0 45252.01 16748.00 10357.77 18146.24 47122.84 34959.38 43126.67
12:34:45 0 60 800 860 860 0 0 45467.35 16963.34 10357.77 18146.24 47122.84 34959.38 43126.67
12:54:56 0 0 800 800 800 0 0 46508.81 18004.80 10357.77 18146.24 47122.84 34959.38 43126.67
15:23:12 0 0 0 0 0 0 0 53625.61 25121.60 10357.77 18146.24 47122.84 34959.38 43126.67
17:23:45 0 30 0 30 30 0 0 53625.61 25121.60 10357.77 18146.24 47122.84 34959.38 43126.67
:00:00 0 30 0 30 0 30 0 53690.86 25186.85 10357.77 18146.24 47122.84 34959.38 43126.67
20:23:23 0 0 0 0 0 0 0 53948.95 25186.85 10615.86 18146.24 47051.15 34959.38 39542.08
Power saved of fuel cell is: 71.80%
Table VII. Simulation results for the Case II. Only fuel cell is depleted.
Cap._left in FC I(Lap) I(GPS) I(PSP) I(tot) I(SC) I(FC) I(Battery) Ch(SC) Ch(FC) Ch(Battery) Cap._left_FC Cap._left_Battery Time Ch(tot) ( C ) without EM (mA) (mA) (mA) (mA) (mA) (mA) (mA) ( C ) ( C ) ( C ) (mA.h) (mA.h) (mA.h) 02:34:56 0 60 0 60 0 0 60 0 0 0 0 2000 40000 2000 03:34:56 2000 60 0 2060 0 0 2060 216.00 0 0 216 2000 39940 2000 06:00:00 2000 60 0 2060 1000 1060 0 18146.24 0 0 18146.24 2000 34959.38 2000 06:34:56 2000 60 10 2070 1000 1070 0 22464.00 2096.00 2221.76 18146.24 1382.84 34959.38 1126.67 07:52:28 2000 60 10 2070 1000 0 1070 32093.64 6748.00 7199.40 18146.24 0.17 34959.38 -811.67 08:34:56 0 60 10 70 70 0 0 37368.00 9296.00 7199.40 20872.6 0.17 34202.06 -811.67 64 09:34:56 1000 60 10 1070 1000 0 70 37620.00 9548.00 7199.40 20872.6 0.17 34202.06 -811.67 09:34:57 1000 60 0 1060 1000 0 60 37621.07 9549.00 7199.40 20872.67 0.17 34202.04 -811.67 11:34:56 0 60 0 60 60 0 0 45252.01 16748.00 7199.40 21304.61 0.17 34082.05 -811.67 12:34:45 0 60 800 860 860 0 0 45467.35 16963.34 7199.40 21304.61 0.17 34082.05 -811.67 12:54:56 0 0 800 800 800 0 0 46508.81 18004.80 7199.40 21304.61 0.17 34082.05 -811.67 15:23:12 0 0 0 0 0 0 0 53625.61 25121.60 7199.40 21304.61 0.17 34082.05 -811.67 17:23:45 0 30 0 30 30 0 0 53625.61 25121.60 7199.40 21304.61 0.17 34082.05 -811.67 18:00:00 0 30 0 30 0 30 0 53690.86 25186.85 7199.40 21304.61 0.17 34082.05 -811.67 18:00:01 0 30 0 30 0 0 30 53690.89 25186.85 7199.43 21304.61 0.16 34082.05 -812.08 20:23:23 0 0 0 0 0 0 0 53948.95 25186.85 7199.43 21562.67 0.16 34010.37 -812.08
From this table, we can see that at 7:52:28, the fuel cell is depleted. But without our energy management, the fuel cell is depleted much earlier. Table VIII. Simulation results for the Case III. Only lithium battery is depleted.
Cap._left in I(Lap) I(GPS) I(PSP) I(tot) I(SC) I(FC) I(Battery) Ch(tot) Ch(FC) Ch(Battery) Cap._left_FC Cap._left_Battery Time Ch(SC) ( C ) FC without (mA) (mA) (mA) (mA) (mA) (mA) (mA) ( C ) ( C ) ( C ) (mA.h) (mA.h) EM (mA.h) 02:34:56 0 60 0 60 0 0 60 0.00 0.00 0.00 0.00 50000.00 2000.00 50000.00 03:34:56 2000 60 0 2060 0 0 2060 216.00 0.00 0.00 216.00 50000.00 1940.00 50000.00 04:31:25 2000 60 0 2060 0 0 0 7197.34 0.00 0.00 7197.34 50000.00 0.74 50000.00 06:00:00 2000 60 0 2060 1000 1060 0 18146.24 0.00 0.00 7197.34 50000.00 0.74 50000.00 06:00:01 2000 60 0 2060 1000 1060 0 18148.30 1.00 1.06 7197.34 49999.71 0.74 49999.58 06:34:56 2000 60 10 2070 1000 1070 0 22464.00 2096.00 2221.76 7197.34 49382.84 0.74 49126.67 65 08:34:56 0 60 10 70 70 0 0 37368.00 9296.00 9925.76 7197.34 47242.84 0.74 46126.67 09:34:56 1000 60 10 1070 1000 70 0 37620.00 9548.00 9925.76 7197.34 47242.84 0.74 46126.67 09:34:57 1000 60 0 1060 1000 60 0 37621.07 9549.00 9925.83 7197.34 47242.82 0.74 46126.25 11:34:56 0 60 0 60 60 0 0 45252.01 16748.00 10357.77 7197.34 47122.84 0.74 43126.67 12:34:45 0 60 800 860 860 0 0 45467.35 16963.34 10357.77 7197.34 47122.84 0.74 43126.67 12:54:56 0 0 800 800 800 0 0 46508.81 18004.80 10357.77 7197.34 47122.84 0.74 43126.67 15:23:12 0 0 0 0 0 0 0 53625.61 25121.60 10357.77 7197.34 47122.84 0.74 43126.67 17:23:45 0 30 0 30 30 0 0 53625.61 25121.60 10357.77 7197.34 47122.84 0.74 43126.67 18:00:00 0 30 0 30 0 30 0 53690.86 25186.85 10357.77 7197.34 47122.84 0.74 43126.67 18:00:01 0 30 0 30 0 30 0 53690.89 25186.85 10357.80 7197.34 47122.83 0.74 43126.25 20:23:23 0 0 0 0 0 0 0 53948.95 25186.85 10615.86 7197.34 47051.15 0.74 39542.08
From this table, we can see that the lithium battery is depleted at 4:31:25. If there is no other active sources, the energy user can’t work any more, until there are other active power sources available after 6:00:00.
Table IX. Simulation results for the Case IV. Fuel cell and lithium battery are depleted.
Cap._left in I(Lap) I(GPS) I(PSP) I(tot) I(SC) I(FC) I(Battery) Ch(tot) Ch(FC) Ch(Battery) Cap._left_FC Cap._left_Battery Time Ch(SC) ( C ) FC without (mA) (mA) (mA) (mA) (mA) (mA) (mA) ( C ) ( C ) ( C ) (mA.h) (mA.h) EM (mA.h) 02:34:56 0 60 0 60 0 0 60 0.00 0.00 0.00 0.00 2000.00 3000.00 2000.00 03:34:56 2000 60 0 2060 0 0 2060 216.00 0.00 0.00 216.00 2000.00 2940.00 2000.00 05:00:33 2000 60 0 2060 0 0 0 10798.22 0.00 0.00 10798.22 2000.00 0.49 2000.00 06:00:00 2000 60 0 2060 1000 1060 0 18146.24 0.00 0.00 10798.22 2000.00 0.49 2000.00 06:00:01 2000 60 0 2060 1000 1060 0 18148.30 1.00 1.06 10798.22 1999.71 0.49 1999.58 06:34:56 2000 60 10 2070 1000 1070 0 22464.00 2096.00 2221.76 10798.22 1382.84 0.49 1126.67 07:52:28 2000 60 10 2070 1200 0 0 32093.64 6748.00 7199.40 10798.22 0.17 0.49 -811.67
66 08:34:56 0 60 10 70 70 0 0 37368.00 9805.60 7199.40 10798.22 0.17 0.49 -811.67 09:34:56 1000 60 10 1070 1070 0 0 37620.00 10057.60 7199.40 10798.22 0.17 0.49 -811.67 09:34:57 1000 60 0 1060 1060 0 0 37621.07 10057.67 7199.40 10798.22 0.17 0.49 -811.67 11:34:56 0 60 0 60 60 0 0 45252.01 10489.61 7199.40 10798.22 0.17 0.49 -811.67 12:34:45 0 60 800 860 860 0 0 45467.35 10704.95 7199.40 10798.22 0.17 0.49 -811.67 12:54:56 0 0 800 800 800 0 0 46508.81 11746.41 7199.40 10798.22 0.17 0.49 -811.67 15:23:12 0 0 0 0 0 0 0 53625.61 18863.21 7199.40 10798.22 0.17 0.49 -811.67 17:23:45 0 30 0 30 30 0 0 53625.61 18863.21 7199.40 10798.22 0.17 0.49 -811.67 18:00:00 0 30 0 30 0 30 0 53690.86 18928.46 7199.40 10798.22 0.17 0.49 -811.67 18:00:01 0 30 0 30 0 0 0 53690.89 18928.46 7199.43 10798.22 0.16 0.49 -812.08 20:23:23 0 0 0 0 0 0 0 53948.95 18928.46 7199.43 10798.22 0.16 0.49 -812.08
Fuel cell and lithium battery are all depleted. From 07:52:28, the solar cell will output its maximum current 1200 mA to the energy users. We can also see that, with our energy management, the fuel cell can operate longer time.
Table X. Simulation results for Case V. No energy user is operating in time slot 1
Cap._left in I(Lap) I(GPS) I(PSP) I(tot) I(SC) I(FC) I(Battery) Ch(tot) Ch(FC) Ch(Battery) Cap._left_FC Cap._left_Battery Time Ch(SC) ( C ) FC without (mA) (mA) (mA) (mA) (mA) (mA) (mA) ( C ) ( C ) ( C ) (mA.h) (mA.h) EM (mA.h) 06:00:00 0 0 0 0 0 0 0 0.00 0.00 0.00 0.00 50000.00 40000.00 50000.00 06:34:56 0 0 10 10 10 0 0 0.00 0.00 0.00 0.00 50000.00 40000.00 50000.00 07:34:56 2000 0 10 2010 1000 1010 0 36.00 36.00 0.00 0.00 50000.00 40000.00 50000.00 08:34:56 0 0 10 10 10 0 0 7272.00 3636.00 3636.00 0.00 48990.00 40000.00 48500.00 09:34:56 1000 0 10 1010 1000 10 0 7308.00 3672.00 3636.00 0.00 48990.00 40000.00 48500.00
67 09:34:57 1000 0 0 1000 1000 0 0 7309.01 3673.00 3636.01 0.00 48990.00 40000.00 48499.58 10:34:56 1000 60 0 1060 1000 60 0 10908.01 7272.00 3636.01 0.00 48990.00 40000.00 48499.58 11:34:56 0 60 0 60 60 0 0 14724.01 10872.00 3852.01 0.00 48930.00 40000.00 46999.58 12:34:45 0 60 800 860 860 0 0 14939.35 11087.34 3852.01 0.00 48930.00 40000.00 46999.58 12:54:56 0 0 800 800 800 0 0 15980.81 12128.80 3852.01 0.00 48930.00 40000.00 46999.58 15:23:12 0 0 0 0 0 0 0 23097.61 19245.60 3852.01 0.00 48930.00 40000.00 46999.58 17:23:45 0 30 0 30 30 0 0 23097.61 19245.60 3852.01 0.00 48930.00 40000.00 46999.58 18:00:00 0 30 0 30 0 30 0 23162.86 19310.85 3852.01 0.00 48930.00 40000.00 46999.58 20:23:23 0 0 0 0 0 0 0 23420.95 19310.85 4110.10 0.00 48858.31 40000.00 43415.00
Power saved of fuel cell is: 82.66%.
Table XI. Simulation results for Case VI. No energy user is operating in time slot 2
Cap._left in I(Lap) I(GPS) I(PSP) I(tot) I(SC) I(FC) I(Battery) Ch(tot) Ch(FC) Ch(Battery) Cap._left_FC Cap._left_Battery Time Ch(SC) ( C ) FC without (mA) (mA) (mA) (mA) (mA) (mA) (mA) ( C ) ( C ) ( C ) (mA.h) (mA.h) EM (mA.h) 04:34:56 0 60 0 60 0 0 60 0.00 0.00 0.00 0.00 50000.00 40000.00 50000.00
05:54:56 0 0 0 0 0 0 0 288.00 0.00 0.00 288.00 50000.00 39920.00 50000.00
06:00:00 0 0 0 0 0 0 0 288.00 0.00 0.00 288.00 50000.00 39920.00 50000.00
07:34:56 2000 0 0 2000 1000 1000 0 288.00 0.00 0.00 288.00 50000.00 39920.00 50000.00 68 08:34:56 0 0 0 0 0 0 0 7488.00 3600.00 3600.00 288.00 49000.00 39920.00 48500.00
18:00:00 0 0 0 0 0 0 0 7488.00 3600.00 3600.00 288.00 49000.00 39920.00 48500.00
19:23:45 0 30 0 30 0 30 0 7488.00 3600.00 3600.00 288.00 49000.00 39920.00 48500.00
19:34:45 0 30 800 830 0 830 0 7507.80 3600.00 3619.80 288.00 48994.50 39920.00 48225.00
20:23:12 0 30 0 30 0 30 0 9920.61 3600.00 6032.61 288.00 48324.27 39920.00 47013.75
20:23:23 0 0 0 0 0 0 0 9920.94 3600.00 6032.94 288.00 48324.18 39920.00 47009.17
20:34:56 1000 0 0 1000 0 1000 0 9920.94 3600.00 6032.94 288.00 48324.18 39920.00 47009.17
22:34:56 0 0 0 0 0 0 0 17120.94 3600.00 13232.94 288.00 46324.18 39920.00 44009.17
Power saved of fuel cell is: 38.64%
Table XII. Simulation results for Case VII. No energy user is operating in time slot 3
Cap._left in I(Lap) I(GPS) I(PSP) I(tot) I(SC) I(FC) I(Battery) Ch(tot) Ch(FC) Ch(Battery) Cap._left_FC Cap._left_Battery Time Ch(SC) ( C ) FC without (mA) (mA) (mA) (mA) (mA) (mA) (mA) ( C ) ( C ) ( C ) (mA.h) (mA.h) EM (mA.h) 06:00:00 0 0 0 0 0 0 0 0.00 0.00 0.00 0.00 50000.00 40000.00 50000.00 06:34:56 0 0 10 10 10 0 0 0.00 0.00 0.00 0.00 50000.00 40000.00 50000.00 07:34:56 2000 0 10 2010 1000 1010 0 36.00 36.00 0.00 0.00 50000.00 40000.00 50000.00 08:34:56 0 0 10 10 10 0 0 7272.00 3636.00 3636.00 0.00 48990.00 40000.00 48500.00 09:34:56 1000 0 10 1010 1000 10 0 7308.00 3672.00 3636.00 0.00 48990.00 40000.00 48500.00
69 09:34:57 1000 0 0 1000 1000 0 0 7309.01 3673.00 3636.01 0.00 48990.00 40000.00 48499.58 10:34:56 1000 60 0 1060 1000 60 0 10908.01 7272.00 3636.01 0.00 48990.00 40000.00 48499.58 11:34:56 0 60 0 60 60 0 0 14724.01 10872.00 3852.01 0.00 48930.00 40000.00 46999.58 12:34:45 0 60 800 860 860 0 0 14939.35 11087.34 3852.01 0.00 48930.00 40000.00 46999.58 12:54:56 0 0 800 800 800 0 0 15980.81 12128.80 3852.01 0.00 48930.00 40000.00 46999.58 15:23:12 0 0 0 0 0 0 0 23097.61 19245.60 3852.01 0.00 48930.00 40000.00 46999.58 15:23:45 0 30 0 30 30 0 0 23097.61 19245.60 3852.01 0.00 48930.00 40000.00 46999.58 17:23:23 0 0 0 0 0 0 0 23312.95 19460.94 3852.01 0.00 48930.00 40000.00 46999.58 18:00:00 0 0 0 0 0 0 0 23312.95 19460.94 3852.01 0.00 48930.00 40000.00 46999.58
Power saved of fuel cell is: 64.34%