Download Second Progress Report.pdf

Practical and Economic Feasibility of Recovery of Waste Energy in
Data Centers by the Thermoelectric Effect
by
Lawrence Lewis III
An Engineering Project Submitted
to the Graduate Faculty
of Rensselaer Polytechnic Institute
in Partial Fulfillment of the Requirements
of the degree of
MASTER OF ENGINEERING IN MECHANICAL ENGINEERING
Approved
_________________________________
Sudhangsu Bose, Ph.D., Project Advisor
Rensselaer Polytechnic Institute
Hartford, Connecticut
April, 2012
© Copyright 2012
by
Lawrence Lewis III
All Rights Reserved
ii
Contents
List of Tables.............................................................................................................................iv
List of Figures............................................................................................................................iv
Nomenclature..............................................................................................................................v
Acknowledgement .....................................................................................................................vi
Abstract ....................................................................................................................................vii
1.
Introduction..........................................................................................................................1
1.1. Background ...................................................................................................................1
1.2. Problem Description......................................................................................................3
1.3. Previous Work...............................................................................................................4
2. Methodology........................................................................................................................5
2.1. Fundamentals of Heat Transfer Applied to Thermoelectric Devices...............................6
2.2. Thermoelectric Generator Fundamental Principles.........................................................7
2.3. Thermoelectric Device Basic Configuration ..................................................................9
2.4. Data Center Cooling Configuration .............................................................................11
2.4.1. Convection Cooling ..............................................................................................11
2.4.2. Conduction Cooling..............................................................................................13
2.4.3. Combination Air and Liquid Cooling....................................................................15
2.4.4. Thermoelectric Device Implementation to Cooling Configuration.........................16
2.4.5. Data Center Functional Parameters .......................................................................16
2.5. Thermoelectric Material Selection ...............................................................................22
2.6. Thermoelectric Performance........................................................................................25
2.7. Economic Feasibility of Using Thermoelectric Devices to Supplement Water Cooled
Cabinets ................................................................................................................................25
3. Results ...............................................................................................................................27
4.
Conclusions........................................................................................................................28
5.
References .........................................................................................................................29
iii
List of Tables
Table 1- Cooling Capability of Commercial Water Cooled Electronics Cabinets.......................17
Table 2- CFM values for given temperature differences with a 25 kW heat load .......................18
Table 3- CFM Values for Commercial Water Cooled Electronics Cabinets Air-to-Liquid Heat
Exchangers ...............................................................................................................................18
Table 4- Temperature difference calculated using equation (9) for given heat loads and CFM
values .......................................................................................................................................19
List of Figures
Figure 1- ASHRAE Equipment Heat Load Trends ......................................................................2
Figure 2- Average Data Center Power Allocation for 12 Benchmarked Data Centers [2].............3
Figure 3- Example of Thermoelectric Junction............................................................................7
Figure 4- Thermoelectric Couple with P and N Type Semiconductors.......................................10
Figure 5- Multiple TE Devices Arranged in a Parallel Configuration.........................................10
Figure 6- Closed Loop Forced Convection Cooling of an Electronics Cabinet...........................12
Figure 7- Open Loop Forced Air Convection Cooling of Electronics Cabinets ..........................13
Figure 8- Conduction Cooled Electronics Cabinet Using a Cold Plate.......................................14
Figure 9- Convection Cooled Electronics Cabinet Rejecting Heat to an In-Cabinet Air-to-Water
Heat Exchanger.........................................................................................................................15
Figure 10- HP® Guidelines for Determining Approximate Heat Removal- Heat capacity for one
rack in a single rack configuration for 25°C server intake air.....................................................22
Figure 11- ZT for P-type thermoelectric materials.....................................................................23
Figure 12- ZT for N-type thermoelectric materials ....................................................................24
iv
Nomenclature
Symbol
A
CFM
k
Units
m2
CFM
W/(m·K)
T
TC
TH
ΔT
V
x
Greek
Symbols
!
°C, K
°C, K
°C, K
°C, K
V
m
Temperature
Temperature of cold junction
Temperature of hot junction
Temperature difference
Voltage
distance
Units
Description
ε
η
σ
ρ
V/K
(µV/K)
S/m
kg/m3
Description
Area
Airflow in cubic feet per minute
Thermal conductivity
Seebeck Coefficient
Magnetic field
Efficiency
Electrical conductivity
Density
v
Acknowledgement
This space reserved for acknowledgements at the completion of the project.
vi
Abstract
The thermoelectric effect has been proven as a source of cooling and small power
generation due to a temperature difference between two junctions of a thermoelectric device as
defined by the Peltier-Seebeck effect. Thermoelectric devices have been used for temperature
regulation by operating as a heat pump to maintain computing devices and integrated circuits at
optimum temperatures for improved processing efficiency. Thermoelectrics have also been used
to capture microwatt electrical power from personal computing and other small scale devices by
way of utilizing the waste heat that is rejected through its heat sink. In modern data centers and
server farms, water cooling of electronics has been widely adapted as a more efficient cooling
method than standard air conditioning and ventilation systems due to its vastly larger thermal
capacity. However, even high density electronics cabinets and processing units are low level
heat applications unfit for waste heat recovery by standard thermodynamics cycles and heat
pumps. When applying the thermoelectric effect to the temperature difference between the heat
source of the processing electronics and the heat sink of a water cooling system, potential exists
for practical and economic energy recovery.
This study demonstrates the feasibility of waste energy recovery from high power density
electronics in data centers and server farms by way of the practical and economic application of
thermoelectricity.
An overview of thermoelectricity and the thermoelectric effect is given,
including a review semiconductor materials and electronics cabinet cooling techniques. This
paper describes an investigation into the efficiency of applying thermoelectricity to low
temperature waste heat situations and the comparison to standard thermodynamic cycles.
Conclusions are presented concerning the effectiveness of this application towards waste heat
utilization for power recovery.
vii
1. Introduction
1.1. Background
Traditional thermodynamic steam cycles require large amounts of heat to convert the
working fluid into a vapor form before it is passed through a turbine. That turbine drives a
generator to create electricity. However, in low heat applications, it is inefficient to use a
traditional cycle, such as the Rankine cycle, because there is insufficient heat to properly
convert the working fluid to a vapor. The initial cost of the equipment necessary for a
Rankine cycle, such as pumps and turbines, also makes traditional cycles unappealing due to
the low rate of return. Thermoelectricity, however, requires relatively little upfront cost, but
also does not generate power on the magnitude of the traditional steam cycles. The devices
necessary are also small and the materials are less toxic. Applying the Seebeck, Peltier, and
Thomson effects, thermoelectricity has been used frequently for thermal conditioning of
electronics.
On the other hand, if a thermal gradient can be maintained between two
junctions of a thermoelectric (TE) module, then electrical power can be generated. As
integrated circuits decrease in size yet increase in their power consumption and essentially
power density, traditional cooling systems may become obsolete in their efforts to properly
cool electronics for normal operation. While thermoelectricity has been proven for cooling
uses, opportunities are arising to benefit from power generation. The power solicited may
not be considerable, but waste energy is free, thereby making the potential for economic and
efficient reuse of energy.
The advantages of TE devices to be able to both cool electronics and generate power
can be leveraged to reduce data center operating costs. In Figure 1 below, the equipment
heat loads as provided by the American Society of Heating, Refrigeration, and AirConditioning Engineers, Inc. (ASHRAE) can be seen.
1
Figure 1- ASHRAE Equipment Heat Load Trends
The United States Environmental Protection Agency estimates that data centers and
server farms in the U.S. alone in 2006 represented approximately 1.5 percent of total
electricity consumption, or 61 billion kilowatt-hours (kWh) [1]. This figure is expected to
almost double over a 5-year period, amounting to over $7 billion in total electricity costs.
However, electronics are also increasing in efficiency, thus leading to less losses and better
use of the energy provided. For example, as seen in Figure 2- Average Data Center Power
Allocation for 12 Benchmarked Data Centers [2]Figure 2, in 2010 HVAC cooling and fans
represented approximately 30 percent of data center power consumption.
2
Figure 2- Average Data Center Power Allocation for 12 Benchmarked Data Centers [2]
This percentage may be able to decrease due to alternative cooling methods with increased
efficiency. The emphasis, however, is placed on the servers themselves and the efficiency of
the electronics. Over 40 percent of electrical power consumption in data centers is spent on
servers, and essentially all this power is rejected through waste heat. Since this waste heat is
rejected to ambient and no longer used, it represents energy that is essentially free and can be
potentially used as a green alternative energy source.
1.2. Problem Description
The objective of this project is to analyze the efficiency of applying TE devices to a
data center to assess the practicality and feasibility of a cost effective solution for waste heat
recovery. Data centers produce relatively low levels of waste heat when compared other
applications that may not make waste heat recovery worth implementing. The efficiency in
the waste heat recovery can assess the magnitude of power generation. Material selection is
necessary to optimize the efficiency of the thermoelectric device while maintaining an
economic solution as well as selection of the optimum heat sink to make execution of TE
devices in a data center realistic.
3
1.3. Previous Work
Several investigations and studies have been completed concerning the subject of
materials for TE devices and recovering waste heat as a green energy source. Thermoelectric
technology is used in countless applications to power small electronics, or harness enough
energy from large heat producing sources to power smaller applications, such as using the
waste heat from an automobile to power the internal computers. Most often, TE devices are
investigated for their use as a cooling device by benefiting from the Peltier effect.
An analysis of thermoelectric cooling of personal computer circuits and waste heat
power generation was conducted by C.A. Gould et al. [3]. This was conducted by mounting
TE modules onto personal computer microprocessors and operating the computers over
various levels of processing. For electrical power generation, a TE module was able to
generate a small amount of power in microwatts.
Automobile manufacturers have also investigated the use of TE devices for waste
heat recovery as explained by Meisner [4]. Following combustion, approximately 40 percent
of gasoline is exhausted through waste heat. The auto industry is working to capture some of
that waste heat through TE modules in the exhaust piping in order to power devices that
normally drain battery life. The goal is to redesign the auto electronics to run on battery
power as the primary power source rather than siphoning power generated by the combustion
of gasoline. The role of gasoline combustion would then be to charge the batteries vice
running the car. Thermoelectric modules can be used to supplement the gasoline combustion
for usable energy and therefore increasing fuel efficiency.
Overall, efficiency levels of TE devices are rated much less than conventional heat
engine cycles, thus granting minimal returns for generating electricity. Thus, data center
waste heat recovery has not been extensively documented due to the low temperature of the
heat rejected.
4
2. Methodology
In order to determine the feasibility and economic practicality of using thermoelectricity
for large scale waste heat and energy recover, the following approach will be adopted:
1. Establish and realistic data center based on published data and investigation. Then define
normal operating parameters for operation including, but not limited to, electronics
electric power consumption, electronics heat rejection, electronics operating temperature,
cooling approach, cooling load requirements, and cooling system power consumption.
2. Determine the potential thermal gradient between heat source and heat sink and correlate
to an appropriate thermoelectric module material selection to optimize efficiency with
consideration to cost.
3. Based on cooling load requirements, determine the operating conditions for the cooling
system.
4. Design cooling system loop using traditional heat exchange equipment to reject full
electronics heat load and maintain constant cooling temperature.
Assumption: Cooling by way of the thermoelectric module will be neglected for this
investigation.
5. Analyze, assess, and make recommendations towards whether heat sink is appropriate to
maximize temperature difference and energy recovery.
6. Investigate potential electricity generation based on chosen thermoelectric module and
efficiency calculations, properties of the materials, environmental conditions of the
electronics and cooling system, and figure of merit of the thermoelectric material.
7. Compare and analyze theoretic electricity generation for data center identified in Step 1.
Cost to implement the system and a possible “break even point” or rate of return may be
identified.
5
2.1. Fundamentals of Heat Transfer Applied to
Thermoelectric Devices
The use of thermoelectric generators (TEGs) for the practical use of recovering
waste heat abides by the laws of thermodynamics.
The First Law of Thermodynamics states that in a closed system, energy can neither
be created nor destroyed. However, the energy can be transferred to other forms of energy
such as heat as well as work. TE devices seek to capture the energy that is being transferred
to heat and return it back to a voltage, or essentially power, without the use of conventional
heat engine, or thermodynamic cycle. These cycles, such as the Rankine cycle, convert heat
to work commonly by the use of equipment such as turbines, which then returns the work to
energy by way of a generator.
The TE device applies the Seebeck effect, which transfers the heat to power like a
heat engine does, without the conventional equipment. While TEGs that apply the Seebeck
effect require semi-conductors and are expensive, yet less efficient, they can recover waste
heat from low temperature applications that conventional heat engines cannot.
The Second Law of Thermodynamics, as defined by the Clausius statement, says that
“No process is possible whose sole result is the transfer of heat from a body of lower
temperature to a body of higher temperature.” Essentially, heat flows from a higher
temperature region or body to a lower temperature region or body and it is not possible to
convert heat completely into work. Losses of heat will occur due to inefficiency. A TEG
applies the Seebeck effect by using a temperature difference to create a voltage. This is
captured through the conduction of heat from what is termed the cold side (side facing the
heat source) of the TEG to the hot side (side facing the heat sink).
6
2.2. Thermoelectric Generator Fundamental Principles
The thermoelectric effect that makes TEGs possible can be defined by two processes,
the Peltier and Seebeck effects.
The Peltier effect occurs when a voltage is applied to two conductors made of
different materials. When the voltage is applied, a circuit can be created that allows for
continuous heat transport between the conductor’s junctions (Saied).
The opposite also applies such that a voltage can be generated by applying a
temperature difference to the conductors, which is known as the Seebeck effect. This
temperature gradient causes charge carriers to diffuse through the materials at an atomic level
by way of electron flow. The heat is transferred in the same direction as the charge carrier
flow, from the cold side of the TEG to the hot side.
Figure 3- Example of Thermoelectric Junction
The voltage (V) generated by this effect can be derived from the following equation:
T2
V = ! (# B (T )" # A (T ))dT
T1
(1)
The voltage is commonly reported as micro-volts per temperature difference in
Kelvin. The Seebeck coefficients αA and αB represent the thermopowers of the two semi-
7
conductor metals A and B as a function of the temperatures T1 and T2, which represent the
temperatures of the hot and cold thermal junctions. The thermopower of a material measures
the magnitude of voltage stimulated in response to a temperature difference across that
material.
The Seebeck coefficient is a material parameter that is used to determine the
efficiency of a given thermoelectric material. The larger a thermoelectric voltage can be
induced combined with a higher Seebeck coefficient results in a higher efficiency.
If the temperature difference dT between the two junctions of a material is
negligible, then the Seebeck coefficient of a material can be approximated as follows:
" =!
dV
dT
(2)
The thermoelectric voltage dV is seen between the two junctions.
The Seebeck coefficient can be related to the electric field ! and the temperature
gradient
dT
by the following equation:
dx
"=
!x
(dT / dx )
(3)
Function (1) is non-linear and depends on the conductor’s absolutely temperature,
material, and molecular structure. However, if the coefficients are essentially constant over
the measured thermal gradient, then the voltage generated can be approximated with the
following formula:
V = (S B ! S A )" (T2 ! T1 )
(4)
Thermoelectric materials are gauged by their figure of merit, which represents their
quality of performance and is defined by the following:
Z=
"
!k
(5)
Where ! is the Seebeck coefficient, ! is the electrical resistivity, and k is the
thermal conductivity. Low electrical resistivity and thermal conductivity are necessary for a
high figure of merit. The figure of merit is commonly multiplied by temperature because
both the electrical resistivity and thermal conductivity are temperature dependent, therefore
making the figure of merit temperature dependent. This also provides the dimensionless
8
figure of merit. The figure of merit can also be defined based on the electrical conductivity as
seen here:
ZT =
" 2!T
k
(6)
The maximum efficiency η of a TE device can be defined using the figure of merit.
" max =
TH ! TC 1 + ZT ! 1
T
TH
1 + ZT + C
TH
(7)
2.3. Thermoelectric Device Basic Configuration
The temperature difference applied to the conductors causes charge carriers to
diffuse from the cold side to the hot side of the TEG. By leaving positively charged nuclei to
collect on the cold side of the TEG while the charge carriers move towards the hot side, a
thermoelectric voltage is generated.
Typically, semiconductors are used in TEGs because they can be doped with
additional electrons or electron holes to increase the Seebeck coefficient. Normal metal
conductors have smaller coefficients due to equilibrium of positive and negative charges in
the material that would induce the thermoelectric voltage. A larger amount of charge carriers
on the hot side of the material results in a higher thermoelectric voltage, and hence
semiconductors are optimum for TE devices.
Commonly used semiconductors are N and P types. N type semiconductors can be
doped with excess electrons, which move towards the positive side of the TEG. A P type
semiconductor is a positively charged carrier and is thus doped with holes, which move in a
direction opposite to the flow of electrons. In both semiconductors, the charge carriers direct
the direction of heat flow, which proceeds in the same direction as the charge carrier, but
electrons and holes move in opposite directions.
P and N type material have different figures of merit and are averaged to determine a
material’s overall quality. To maximize the efficiency of a TE device, P and N type
semiconductors are placed in series electrically, but are thermally in parallel, which creates a
couple as seen below in Figure 4.
9
Figure 4- Thermoelectric Couple with P and N Type Semiconductors
As stated previously, the cold side of this couple is facing the heat source, where the
heat is absorbed, and the hot side is where the heat is rejected to the heat sink.
As can be seen in the example couple above, a ceramic substrate exists as an
electrical insulator between the heat source and the conductor as well as between the
conductor and heat sink. This is used to prevent an electrical short circuit between the
conductors and the heat source and sink. Ceramics are commonly used between the electrical
insulator needs a high thermal conductivity to minimize the thermal gradient between the
conductors and the heat source and sink. When multiple TE devices are used in parallel it
resembles the configuration seen in Figure 5.
Figure 5- Multiple TE Devices Arranged in a Parallel Configuration
10
2.4. Data Center Cooling Configuration
Data centers come in wide arrays of functionality. As the power densities of
electronics continue to increase, consequently so does the heat load within data centers. This
has lead to numerous cooling solutions for electronics cooling that are efficient, reliable, and
cost effective. The approaches used for electronics cooling are convection to air, conduction
to liquid, or a combination of the two. The data center facility and impacts of the cooling
system are taken in consideration implementing that cooling system into the facility
infrastructure as well as cooling system to electronics enclosure integration.
2.4.1.
Convection Cooling12
Convection cooling is employed by passing conditioned air over electronics
equipment or components within the electronics enclosure to remove waste heat by
convection. The conditioning of this air can occur both internal and external to the
cabinet. When performing the air conditioning external to the cabinet, two basic
approaches may be utilized.
The first approach is a closed loop forced convection cooling system that
relies on large ventilation components, or computer room air conditioning (CRAC) units,
and a network of supply and return distribution ductwork to circulate the air between the
ventilation components and each electronics enclosure. The air is forced through the
electronic components within the cabinets, thus producing heat transfer via convection as
seen in Figure 8.
1
2
Chapter 4 of Liquid Cooling Guidelines for Datacom Equipment Centers.
American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc. (2006). Liquid Cooling Guidelines for Datacom Equipment Centers,
Atlanta.
11
Figure 6- Closed Loop Forced Convection Cooling of an Electronics Cabinet
This air is then vented to the ambient environment or directly ducted away
from the electronics where it is conditioned again by way of a chiller, and eventually the
heat is rejected to the ambient environment outside of the data center facility. As heat
loads increase, this approach requires substantial arrangement volume for large
ventilation components that would be necessary to penetrate the various electronics
enclosures and distribution ductwork.
The second approach is an open loop forced convection cooling system that
relies on fans within the cabinets to draw ambient air over the electronic components to
remove heat through convection. The heated air is discharged back to the ambient
environment. This heated air is then conditioned via the facilities ventilation system’s
chillers. This approach usually dictates an isolated electronics area with a lower ambient
temperature that is below the acceptable ambient air temperature range for manned areas.
12
Figure 7- Open Loop Forced Air Convection Cooling of Electronics Cabinets
Separation of personnel from the electronics equipment provides an
opportunity to maintain ambient temperatures within the electronics area at these reduced
temperatures by way of thermally isolating the equipment areas. However, this also
requires maintaining the facility space within temperature tolerances and may encounter
issues with temperature stability. Data centers have seen “hot spots” of air temperatures
exceeding the cooling air inlet temperature requirements, which may impact reliability
and performance of the electronics.
Convection cooling by air is the most common approach to forced cooling due
to a lower relative cost and easier implementation than liquid cooling. However, air also
has a low thermal capacity, thereby making it less efficient than liquids for heat removal
as heat loads increase. Also, fan assemblies make
2.4.2.
Conduction Cooling
Liquid cooling is generally employed for high heat load applications since
liquids such as water have a thermal capacity of approximately 3,500 times larger than
air. This translates into the heat source rejecting more heat into the liquid cooled heat
sink, and the temperature increase of the liquid will be less than that of water.
13
Per the ASHRAE3, liquid cooling has been implemented if liquid is circulated
to and from the electronics enclosure for operation. This is commonly done through two
closed liquid cooling loops. One cooling loop provides the liquid directly within the
electronics enclosures for waste heat removal and rejects the waste heat to the second
loop, which is cooling liquid supplied to data center by the facility. The facility provided
liquid cooling loop is known as a primary loop, whereas the loop interfacing with the
electronics is a passive secondary loop that just circulates liquids through heat
exchangers. This secondary loop circulates a conditioned liquid to meet the electronics
requirements. Besides being conditioned to temperature requirements, this may also
include quality such as contaminant levels, filtration levels, and conductivity levels. Fluid
choice can range from water (potentially deionized) to refrigerants and dielectric fluids.
Depending on environmental conditions, particularly data center operating dew point,
cooling water commonly interfaces with base of the electronics enclosures due to
condensation or liquid leakage concerns.
When employing only liquid conduction cooling, cooling liquid is pumped
through devices such as cold plates within the enclosures to which heat producing
components are mounted and through which heat is conducted into the liquid, as seen in
Figure 8. This heat is subsequently transferred to the facility primary cooling loop via
liquid to liquid heat exchangers external to the electronics cabinets, and ultimately
rejected to ambient outside of the facility.
Figure 8- Conduction Cooled Electronics Cabinet Using a Cold Plate
3
American Society of Heating, Refrigerating and Air-Conditioning Engineers, Inc. (2006). Liquid Cooling Guidelines for Datacom Equipment Centers,
Atlanta.
14
2.4.3.
Combination Air and Liquid Cooling
The most common approach to liquid cooling of electronics enclosures is to
implement an air to liquid heat exchanger within the cabinet whereby all waste heat is
rejected from the enclosures through the cooling liquid. This utilizes a combination of
both air and water for waste heat removal. The electronics are cooled as identified above
by passing conditioned air over the components. The air is then conditioned internal to
the cabinet by way of the air to liquid heat exchanger, which is made up of a compact
cooling coil and fan assembly. This assembly can be located in multiple configurations
within the enclosure, but consideration should be taken to avoid dripping onto the
electronics components in the event of condensation or leakage as stated above. This
assembly is most often used in an arrangement within the enclosure that directs the
cooling air in a loop and thus does not mix the hot and cold air streams. This maximizes
heat removal and decreases temperature stability issues common with the open loop
forced convection approach identified above.
Figure 9- Convection Cooled Electronics Cabinet Rejecting Heat to an In-Cabinet Air-to-Water Heat
Exchanger
While numerous other configurations exist utilizing both air and liquid as
mediums for waste heat removal, this study is limited to the conventional methods of
waste heat removal.
15
2.4.4.
Thermoelectric Device Implementation to Cooling
Configuration
In order to maximize TE device implementation, a closed cabinet liquid
cooling system is considered optimum. A localized cooling system internal to the
electronics enclosures provides thermal junctions that the TE devices can be attached to
while minimizing contact resistance. Also, by using a closed system where all waste heat
is rejected through a single source that has a consolidated footprint, the temperature
difference can be maximized while minimizing losses of waste heat to ambient.
Accordingly, all energy entering the cabinet is assumed to be discarded through waste
heat.
Liquid cooling systems are also more efficient at waste heat removal for large
heat load applications. As the power density of electronics continues to increase, liquid
cooling allows for future growth in data center functionality as well as decreasing
operating costs.
2.4.5.
Data Center Functional Parameters
2.4.5.1.
Data Center Air Temperature
Per ASHRAE, data centers are typically considered to be under Class A1
due to commonly having tightly controlled environmental parameters. These
parameters include the operating temperature, relative humidity and consequently
dew point. They are also considered to be involved with mission critical operations.
While some data centers may not fall into this category, Class A1 is used in this study
to represent the typical data center environmental requirements. However, ASHRAE
guidelines are intended as recommendations rather than limitations. Regardless, Class
A1 parameters are used.
As such, this study is interested in the normal operating temperature band
of the typical data center. As seen in the ASHRAE Class A1 guidelines, the
operational temperature is specified as 18°-27°C (64.4°-80.6°F) or 15°-32°C (59°89.6°F) allowable. To maximize the temperature difference between the junctions of
a TEG, the upper limit of the electronics cooling air temperature is set at 27°C
(80.6°F).
16
2.4.5.2.
Data Center Cabinet Heat Load
An investigation of known data center operators and electronics cabinet
manufacturers was completed to determine a nominal heat load value for a single
typical water cooling cabinet. After inspection of water cooled units by APC®, Black
Box Networking Solutions®, HP®, Intel®, and Rittal®, the cooling capacity of the
single cabinets is seen to range from 20kW to 37 kW for standard level high density
electronics cabinets with negligible heat losses to ambient for each unit.
Table 1- Cooling Capability of Commercial Water Cooled Electronics Cabinets
Manufacturer
APC® InRow Direct Expansion Cabinets
Black Box Networking Solutions® EliteTM
Cabinets with Coolant Management System
Hewlett Packard® Modular Cooling System
kW
37
Intel®4
24
Rittal® Liquid Cooling Package Standard
Modules
20
33
35
Electronics cabinet manufacturers are constantly upgrading their
technology to meet the cooling demands of modern electronics. Numerous other
manufacturers also provide solutions to supplement current cabinet technology to
improve the cooling capacity. For this project, the standard heat load of a single
electronics cabinet is defined conservatively at 25 kW based on the above values.
2.4.5.3.
Data Center Air Flow Rate
When employing air-to-liquid heat exchangers internal to the electronics
cabinets, the ability of the heat exchanger to effectively cool the electronics is
dependent on the volume of air throughput expressed in cubic feet per minute (CFM).
By increasing the airflow across the electronics per a given time, the heat removal
capacity increases. This is primarily controlled by the use of fans that supplement
cooling coils in a heat exchanger. The size of the fans used directly correlates to
4
Based on Intel® Information Technology Brief “Expanding Data Center Capacity with Water-Cooled Cabinets”
from June 2007
17
CFM; the larger the fan, the more airflow. However, as fan size, revolutions per
minute (RPM), and the number of fan blades increase, so does the cost associated
with the unit and vibration and acoustic noise levels. Also, fans have limits, because
even as they increase in CFM, they cannot cool lower than the surrounding
temperature [5].
One way to determine the optimum amount of CFM necessary for the
cooling load is to use the following formula that correlates the heat load, airflow, and
temperature difference between intake and exhaust air temperatures in °F.
Watts = .316 " CFM " !T
(8)
This can be rearranged to solve for CFM:
CFM =
Watts
.316 " !T
(9)
Typical data centers use a !T value of 20°-30°F. Applying the electronics
heat load defined above of 25 kW to formula (9), the corresponding CFM values for
each of these temperature differences is as follows:
Table 2- CFM values for given temperature differences with a 25 kW heat load
CFM
3956
2637
!T
20°F
30°F
For comparison, the CFM values used by air-to-liquid heat exchangers for
the water cooled cabinet manufacturers identified in Table 1 are listed in Table 3.
Table 3- CFM Values for Commercial Water Cooled Electronics Cabinets Air-to-Liquid Heat Exchangers
Manufacturer
APC® InRow Direct Expansion Cabinets
Black Box Networking Solutions® EliteTM
Cabinets with Coolant Management System
Hewlett Packard® Modular Cooling System
5
Intel®
CFM
2290
2500
2500
800-1200
Rittal® Liquid Cooling Package Standard
Modules
5
Up to 1412
Based on Intel® White Paper “The State of Data Center Cooling” by Michael K. Patterson and Dave Fenwick,
March 2008
18
The CFM values in Table 3, when related to their respective heat loads in
Table 1, would require !T values far in excess of 40°F as calculated and displayed in
Table 4.
Table 4- Temperature difference calculated using equation (9) for given heat loads and CFM values
Manufacturer
APC® InRow Direct Expansion Cabinets
Black Box Networking Solutions® EliteTM
Cabinets with Coolant Management System
Hewlett Packard® Modular Cooling System
!T (°F)
51.1
41.8
44.3
Intel®
63.3
Rittal® Liquid Cooling Package Standard
Modules
44.8
These CFM values suggest that the electronics cooling systems for the
listed cabinet manufacturers have cooling capabilities above and beyond what is
suggested through equation (9).
Thus, based on the CFM calculated for a 30°F temperature difference
using equation (9), and the values in Table 3, the in cabinet airflow for this project is
set at 2500 CFM.
2.4.5.4.
Data Center Cooling Liquid Selection
Fluid selection for liquid cooling application can have a large impact, but
the liquid used for the cabinet cooling loop is not of utmost importance for this
project due to the focus on operating temperature rather than liquid performance. The
objective is to optimize the !T across the TE module in order to maximize power
generation.
However, the liquid does dictate both economic and environmental
impacts.
Common coolant selections for liquid cooling applications are water,
deionized water, ethylene glycol, and other dielectric fluids. The coolant selection for
this application is a product of thermal needs, such as thermal conductivity and heat
capacity. Consequently, Section 2.4.5.1 and Section 2.4.5.3 have established both the
cabinet operating air temperature and the appropriate temperature difference between
the intake and exhaust air flowing through the cabinet. This provides an operational
temperature window for the coolant.
19
Glycol solutions have beneficial thermal properties such as a low freezing
point, which can increase the temperature difference across the TE module. It also
has a high thermal conductivity and specific heat, which makes it ideal for this
application. However, glycol solutions do not offer as high a thermal conductivity as
water, and introduce toxicity or environmental concerns by dealing with such fluids.
The benefit of glycol is to offer freeze protection, but typical data centers do not
control their dew point to such low levels that excessive condensation may become a
concern. Thus, glycol solutions are not considered for this project.
Dielectric fluids are popular for electronics industry cooling systems
because of their non-conductive properties that could be a concern to sensitive
electronics. Like glycol solutions, its benefit is that it can operate over a wide
temperature range. However, the thermal conductivity of dielectric fluids is at least
four times less than that of water, and therefore is considered for this project.
Water offers high thermal conductivity and heat capacity levels for this
cooling application. The conductive and corrosive properties of the water can change
depending on whether the end user desires deionized water. Deionized water is
commonly preferred for electronics cooling applications because it is non-conductive
and acts an insulator to electronics. This may increase the cost of the system by
introducing filtration standards and necessitating higher quality piping materials to
resist corrosion. However, the change in thermal properties is negligible. Whether
deionized or not, water offers a low cost cooling solution as long as corrosion is
monitored. Due to the low environmental impact, large thermal conductivity, and
high heat capacity, water is appropriately selected as the coolant for the liquid cooling
system.
2.4.5.5.
Data Center Chilled Water Temperature
The operational temperature for the data center secondary cooling loop is
an important balance between cooling system efficiency and TE module performance.
The objective of this project is to reduce energy use, therefore emphasis is placed on
the efficiency of the chilled water system secondary to primary loop heat exchanger.
Per U.S. Department of Energy Best Practices [6], high efficiency chilled water
20
systems operate with a medium temperature chilled water.
This translates to a
cooling water temperature of 55°F or higher. The benefit to using a higher cooling
water temperature, in addition to chiller efficiency, is a reduced risk of condensation.
Operating above the data center ambient air dew point avoids the necessity for
dehumidification and conditioning of ambient air.
Otherwise, condensation
mitigation practices must be implemented, such as condensation draining and
ruggedizing of components including the cabinet heat exchangers, TE modules, and
potentially the electronics being cooled. This medium level temperature also assists
cooling system design when implementing economizers, which can be employed by
the data center but are outside the scope of this project.
Thus, for this project, 55°F will be adopted for the chilled water operating
temperature. This is used as a guideline and may be deviated from to meet specific
user demands.
2.4.5.6.
Data Center Chilled Water Flow Rate
Chilled water flow rates are dependent upon many factors. Arrangement
impacts are of particular concern to electronics cabinets manufacturers as much as
temperature. Higher flow-to-load ratios minimize cabinet heat exchanger size but
adversely impact the facility water distribution equipment, such as increasing the size
of pumps and distribution piping. In this study, initial water temperature is the only
concern, thus the flow rate will be selected based on existing water cooling systems.
Guidelines from HP® for their single water cooled server rack configuration are seen
below.
21
Figure 10- HP® Guidelines for Determining Approximate Heat Removal- Heat capacity for one rack in a
single rack configuration for 25°C server intake air6
Based on Figure 10, and the cabinet heat load selection in Section 2.4.5.2,
the chilled water flow rate selected for the cooling of a single electronics cabinet is
approximately 10 gpm when using 12.5°C (54.5°F) chilled water.
2.5. Thermoelectric Material Selection
Based on the data center functional parameters defined above in Section 2.4.5, the
materials for the TE device can be determined. The heat transfer equation seen below can be
used to calculate the temperature of the electronics exhaust air.
(8)
Q = m& c P !T
Where Q represents the cabinet heat load, m& is the air flow rate, and !T represents
the temperature difference between the exhaust temperature entering the air-to-water heat
exchanger and the electronics intake air temperature.
Rearranging equation (8) and
substituting for !T , the exhaust air temperature, or essentially the hot junction temperature,
is calculated here.
6
Based on “HP Modular Cooling System Generation 2 Site Preparation Guide” from February 2008
22
TH =
Q
+ TC
m& c P
(9)
Where m& for air can be defined as:
m& = V&!
(10)
For this temperature level, the most commonly used semiconductor material choice
is Bismuth Telluride (Bi2Te3). This material has a high figure of merit for the relatively low
operating temperature range applicable to electronics cooling. Figure 11 shows a chart of the
dimensionless Figure of Merit for various P-type semiconductors over a wide temperature
range.
Figure 11- ZT for P-type thermoelectric materials7
As can be seen, Bi2Te3 has the highest Figure of Merit compared to other materials in
the lower temperature range. Likewise, when charting N-type semiconductors in Figure 11,
Bi2Te3 still has the highest Figure of Merit for the applicable temperature range.
7
(Snyder, J. http://www.its.caltech.edu/~jsnyder/thermoelectrics/science_page.htm)
23
Figure 12- ZT for N-type thermoelectric materials8
Below is a plot of the figure of merit (Z), Seebeck coefficient, electrical resistivity, and thermal
conductivity, as a function of temperature for Bi2Te3. Carrier concentration will alter the values
below.
8
(Snyder, J. http://www.its.caltech.edu/~jsnyder/thermoelectrics/science_page.htm)
24
(Yazawa, 2005)
Metals are used to sandwich the semiconductor. Because the TE performance is also dependent
on these materials, an optimal material must be chosen, usually copper.
2.6. Thermoelectric Performance
TE performance depends on the following factors:
The temperature of the cold and hot sides.
Thermal and electrical conductivities of the device’s materials.
Contact resistance between the TE device and heat source/heat sink.
Thermal resistance of the heat sink.
2.7. Economic Feasibility of Using Thermoelectric
Devices to Supplement Water Cooled Cabinets
25
26
3. Results
This section is reserved for results of this study.
27
4. Conclusions
This section is reserved for conclusions at the completion of analysis.
28
5. References
The following published works will be used as guidance and references in performing this study:
[1] U.S. Environmental Protection Agency ENERGY STAR Program. Report to Congress on
Server and Data Center Energy Efficiency Public Law 109-43. 2 August 2007.
[2] Koomey, Jonathan G., Ph.D. Estimating Total Power Consumption by Servers in the U.S. and
the World. Lawrence Berkeley National Laboratory. 15 February 2007.
[3] Gould, C.A.; Shammas, N.Y.A.; Gainger, S.; Taylor, I.; Thermoelectric cooling of
microelectronic circuits and waste heat electrical power generation in a desktop personal
computer. Staffordshire University. Beaconside, Stafford, UK. Elsevier B.V. 2010.
[4] Meisner, Gregory P. Advanced Thermoelectric Materials and Generator Technology for
AutomotiveWaste Heat at GM. General Motors Global Research and Development. Warren, MI.
January 2011.
[5] Pentair Technical Products. Overview & Design of Data Center Cabinets. Anoka, Minnesota.
2010.
[2] Dresselhaus, Mildred et al. Perspectives on Recent Advances in Thermoelectric Materials
Research. Massachusetts Institute of Technology. Cambridge, MA. 30 September 2009.
[3] Energy Information Administration. U.S. Household Electricity Report. National Energy
Information Center, 14 July 2005. http://www.eia.doe.gov/emeu/reps/enduse/er01_us.html
[5] Goupil, Christophe; Seifert, Wolfgang; Zabrocki, Knud; Mueller, Eckhard; Snyder, G.
Jeffrey. Thermodynamics of Thermoelectric Phenomena and Applications. Laboratoire
CRISMAT. Caen, France. 15 August 2011. ISSN 1099-4300.
[6] Greenwood, Veronique. Will Your Next Furnace Be A Server Farm? Discovery Magazine, 26
July 2011.
http://blogs.discovermagazine.com/80beats/2011/07/26/will-your-next-furnace-be-a-server-farm/
[7] Miller, Rich. Google’s Energy Story: High Efficiency, Huge Scale. DataCenterKnowledge, 8
September 2011.
http://www.datacenterknowledge.com/archives/2011/09/08/googles-energy-story-highefficiencyhuge-scale/
[8] Niccolai, James. Greener solutions for data centres. Techworld, 11 September 2009.
http://news.techworld.com/green-it/3201589/engineers-boost-server-farm-cooling-efficiency/
29
[9] Rowe, David Michael. Thermoelectric Waste Heat Recovery as a Renewable Energy Source.
Cardiff University. Cardiff, Wales, UK. International Journal of Innovations in Energy Systems
and Power, Vol. 1, no. 1, November 2006.
[10] Saied, Widah. Essentials of Thermoelectric (TE) Cooling. With an Emphasis in Thermal
Control of Electronics.
[11] Teschler, Leland E. Waste heat: Best green power source? Machine Design, 4 May 2011.
http://machinedesign.com/article/waste-heat-best-green-power-source-0504
30