Simulation of a novel copper heat Sink using copper pipe and AM method for CPU group heat removing in power transformer’s cabinet Jafar Mahmoudi2 Kourosh Mousavi Takmai1 1. Ph.D. student in Malardalen University, 2. Professor in Malardalen University IST dep., M’lardalen university, Box 883, Se721 23, V’ster[s, sweden
[email protected] [email protected]
Abstract: Heat sinks operate by conducting heat from the processor to the heat sink and then radiating it to the air. The better the transfer of heat between the two surfaces (the CPU and the heat sink metal) the better the cooling. Some processors come with heat sinks glued to them directly, ensuring a good transfer of heat between the processor and the heat sink. In this paper author have simulated a new copper heat sink and heat pipe (is a simple device that can quickly transfer heat from one point to another) that has a best heat transferring. A three Dimensional finite element is used for simulations of temperature behaviour on around of heat sink. Analytically approach is applied to determine of heat transfer coefficients. The method has a good convergence and is adaptive with other best designed heat sinks. And so we examine the use of activity migration which reduces peak junction temperature by moving computation between multiple replicated units. Introduction Power transformers are vital devices in power transmission networks. Many CPU, chips and other electronic and power electronic devices are used in control and protection of power transformers, same as relays, control cabinet on the tank of transformers or control room cabinets and so in auxiliary systems of power transformers. In warm days or in overloading conditions, the utilities need to a reliability working of power transformers, then correct performance of electronic device due to ambient high temperature or other conditions are necessary. Relays use many chips or cpu and they has a important role on protection or controls. Power dissipation is emerging as a key constraint on achievable processor performance. Elevated die temperatures reduce device reliability and transistor speed, and increase leakage currents exponentially. Providing adequate heat removal with low cost packaging is particularly challenging as processor power densities rise above 100 W/cm2. Moreover, power dissipation is unevenly distributed across a microprocessor die. Highly active portions of the design such as the issue window or execution units may have more than twenty times the power density of less active blocks such as a secondary cache [5]. Even with dynamic thermal management, these hot spots limit performance because total power dissipation must be reduced until all hot spots have acceptable junction temperatures. In this paper, have used heat pipe system and so AM models, we evaluate the use of activity migration to reduce power densities. Hot spots develop because silicon is a
relatively poor heat conductor and cannot spread heat efficiently across a die. With activity migration (AM), we instead spread heat by transporting computation to a different location on the die. Computation proceeds in one unit until it heats past a temperature threshold. The computation is then transferred to a second unit allowing the first to cool down. We show how AM can be used either to double the power that can be dissipated by a given package, or to lower the operating temperature or hence the operating power. Our thermal model predicts that the intervals between migrations needed to obtain maximum benefit can be much smaller than typical operating system context swap times, and will scale to smaller intervals in future technologies. We evaluate the performance impact of such small migration intervals, and examine various alternative architectural configurations that trade area for power/performance. Heatsink materials The thermal conductivity of the heatsink's material has a major impact on cooling performance. Thermal conductivity is measured in W/mK; higher values mean better conductivity. As a rule of thumb, materials with a high electrical conductivity also have a high thermal conductivity. Alloys have lower thermal conductivity than pure metals, but may have better mechanical or chemical (corrosion) properties. The following materials are commonly used for heatsinks: - Aluminum: It has a thermal conductivity of 205W/mK, which is good (as a comparison: steel has about 50W/mK). The production of aluminum heatsinks is inexpensive; they can be made using extrusion Due to its softness, aluminum can also be milled quickly; die-casting and even cold forging are also possible (see part 2 of this guide for more information about production methods). Aluminum is also very light (thus, an aluminum heatsink will put less stress on its mounting when the unit is moved around). - Copper's thermal conductivity is about twice as high as aluminum - almost 400W/mK. This makes it an excellent material for heatsinks; but its disadvantages include high weight, high price, and less choice as far as production methods are concerned. Copper heatsinks can be milled, diecast, or made of copper plates bonded together; extrusion is not possible. To combine the advantages of aluminum and copper, heat sinks can be made of aluminum and copper bonded together. Here, the area in contact with the heat source is made of copper, which helps lead the heat away to the outer parts of the heatsink. The first heatsink for PC CPUs with an embedded copper piece was the Alpha P7125 (for firstgeneration Slot A Athlon CPUs). Keep in mind that a copper
embedding is only useful if it is tightly bonded to the aluminum part for good thermal transfer. This is not always the case, especially not with inexpensive coolers. If the thermal transfer between the copper and the aluminum is poor, the copper embedding may do more harm than good; The copper plate helps spread heat across the base plate. Thermal right heat sink (prototype) with large heat pipe in the center, a heat pipe provides substantially better thermal transfer than a solid piece of copper. - Silver has an even higher thermal conductivity than copper, but only by about 10%. This does not justify the much higher price for heatsink production - however, pulverized silver is a common ingredient in high-end thermal compounds
needed, just enough to fill the gap between the CPU and heat sink. Using more will not make it work better, it will just make a big mess when you press the heat sink down onto the CPU.
Semiconductor Materials and Wafer Manufacture Processors are manufactured from semiconductor material. Semiconductors are materials that transmit electricity only under certain conditions, and therefore are ideal for making the ultra-small, high-speed transistors that implement the modern processor. Only certain materials are suitable for use in semiconductors. To be suitable, the material must be (at the very least): Equivalent RC Thermal Model • Able to be made into a semiconductor. A chemical process called "doping" is used to introduce small impurities into a pure material, which enables the material to act electrically as a semiconductor. • Able to be manufactured into large pieces of uniform composition and high quality. t Rsilicon, vertical = • Of the appropriate hardness so that it can be cut into k × Ablock Figure1. Equivalent RC Thermal Model t Adie thin slices without being so brittle that it cracks. Rpackage, vertical = 120 × × k Ablock • Readily available and relatively inexpensive, to t Rtotal , vertical = (1 + 120 × Adie) × ensure supply and keep costs down. k × Ablock Csilicon = c × t × Ablock The most famous, and widely used material for Equivalent RC Thermal Model is consisting of: semiconductors is of course silicon. It is used for processors, temperature - voltage, power – current elements. memory, and most other mainstream chips. Another popular Thermal resistance: lateral resistance ignored material for use in semiconductors is gallium arsenide Thermal capacitance: package capacitance modeled as a (GaAs), which is not nearly as widely encountered, and is temperature source (isothermal point) generally used for specialty applications. Exponential dependence of leakage power on temperature Silicon has several great advantages that make it ideal for modeled as voltage-dependent current source (P_leakage(Tj)) use for semiconductors. Obviously it is a semiconductive Junction temperature is modeled by a simple first-order material. It also has several manufacturing advantages that equivalent RC circuit using the well-known analogy between make it the material of choice: it is plentiful and hence cheap electrical circuits and thermal models. Temperature and power in its raw form (regular sand is silicon dioxide). But most are represented by voltage and current respectively. importantly, it can be grown into large, uniform crystals. The Only vertical thermal resistances are considered since first step in the manufacture of a chip is growing these large most heat is dissipated vertically in a chip particularly when crystals of silicon. This is similar in concept to how you can wafers are thin. The heat spreading by the heat-sink is grow sugar crystals in a cup of water, although of course, accounted for by the selection of the isothermal point. special techniques are used. Only thermal capacitance of die is considered since These crystals are very large--the common current size is package capacitance is so massive that it acts as a temperature 8 inches in diameter, and this will probably actually increase source or isothermal point in the figure. to 12 inches very soon!; The larger the crystal, the more chips Exponential dependence of leakage power on temperature that can be manufactured at the same time, and the less waste material, thus saving money. Despite being made from a is modeled as voltage-dependent current source. cheap raw material, these crystals are expensive due to the precise technology used to make them, and the extremely high Heat sink compound Heat sinks that are attached using clips normally sit rather quality standards required to make a crystal suitable for use in loosely on top of the processor. It may feel like it is attached chip making. A single, high-quality crystal will be used to securely, but there will be a gap between the CPU and the make many thousands of processors. These large crystals are cut into wafers by high-precision heat sink, and that gap of air them makes for poor heat transfer, even if it is very small. Air is a poor conductor of saws. These thin slices of semiconductor material then heat compared to most liquids or solids. To improve the undergo the process that transforms them into actual thermal connection between the processor and heat sink, a integrated circuits. Each crystal is cut into several hundred special chemical called heat sink compound should be used. A wafers, each less than 1 mm thick (one fortieth of an inch) and thin layer of this is spread between the two, which greatly 8 inches in diameter. The thinner the wafers, the more that can be made from a single crystal, but the more fragile they are improves heat transfer and the cooling of the processor. Heat sink compound is typically a white paste made from and the harder they are to manipulate. The wafers undergo zinc oxide in a silicone base. Very little of the substance is several more steps before being ready for use, including precision polishing and chemical treatment.
The days where heat pipes were mostly used for temperature equalization in spacecrafts and satellites are over. Nowadays, we commonly find heat pipes in notebook computers, game consoles, and even integrated into normal PC CPU coolers. One reason for the rise in popularity is the fact that prices have dropped dramatically, since high-volume cooling product manufacturers like Asia Vital Components now have their own heat pipe manufacturing facilities, and heat pipe manufacturing is no longer reserved to a few specialized companies. When it comes to PC cooling, "heat pipe" has become a buzzword; but still, few people understand how heat pipes work, and what factors must be considered when using a heat pipe-based cooling system. This article should provide some clarifications. Hot Spots Due to hot spot in cpu: • Rapid rise of processor power density • Uneven distribution of power dissipation • Blocks such as issue windows have more than 20xpower density of less active block such as L2$ • Reduced device reliability and speed, increased leakage current Existing Solutions for hot spot removing are: Packaging/cooling but it has high cost, Dynamic thermal management and it is performance loss, Total power dissipation must be reduced until all hot spots have acceptable junction temperature. Heat pipes A heat pipe is a device has an extremely high thermal conductivity, and is used to transport heat. In order to achieve this, heat pipes take advantage of simple physical effects: As a liquid evaporates, energy - in the form of heat - must be taken from the environment. Therefore, an evaporating liquid will cool the surrounding area. This is how a heat pipe effectively cools the heat source. However, this doesn't get rid of the heat; heat is just transported with the vapor. At the target side for heat transport, the heat pipe must be cooled, for example using a heat sink. Here, the inverse effect takes place: The liquid condenses, and therefore emits heat. Using these effects, it is possible to build heat pipes that have a thermal conductivity that is many thousand times higher than a copper piece of the same size. Note that unlike Peltier elements, a heat pipe does not consume energy or produce heat itself. It is also not possible to cool a device below ambient temperature using a heat pipe.
Figure2. Heat pipe made by Thermacore
Design of heat pipes, and heat pipe related problems In order to understand heat pipe-related problems, we will look at a simple model heat pipe, which is actually more of a thermo syphon. Following the explanations in the above paragraph, you can imagine a simple device: A closed metal tube in vertical orientation, filled with a small amount of water, and a heatsink mounted on its upper side. If you apply a heat source to the lower side, the water will evaporate, while cooling the heat source. The vapor will move up, and condense near the heatsink on top. From there, water will drip back to to the bottom, evaporate again, etc. This simple model already shows the common problems related to heat pipe technology: At normal atmospheric pressure, this would allow the heat pipe to keep the heat source at a temperature of around 100 degrees Celsius, the temperature at which water evaporates. However, since heat pipes are tightly sealed tubes, the exact evaporation temperature can be further adjusted to a certain extent by varying the pressure inside the tube (usually reducing it). To cover other temperature ranges, heat pipe manufacturers use various working fluids instead of water, to cover a very wide range of temperatures. One problem stays: A heat pipe designed for a certain temperature range will only work well in this range; designing a universal heat pipe for all temperature ranges is not possible. Apart from the vapour temperature range, factors like thermal stability and thermal conductivity influence the choice of working fluid. Suitable working fluids include: • For ultra low temperatures: inert gases (helium), nitrogen, ammonia • For usual temperatures to meet electronics cooling requirements: Distilled water with various additions, organic substances like acetone (think nailpolish remover), methanol, ethanol (think booze), toluene (think magic markers). • For high temperatures: Metals like mercury, sodium, silver. Vertical orientation required - In our simple model heat pipe, the working fluid simply drips back to the heat source. It is quite obvious that this design will only work in vertical orientation. To overcome this limitation, commercially available heat pipes do not rely on gravity alone to move the liquid back to the heat source; they take advantage of capillary action. The inside of the heat pipe tube is filled with a capillary structure, often referred to as wick. Different structures are being used: • The most simple structure is a grooved tube. Here, capillary action merely helps gravity; vertical orientation of the heat pipe is usually still required. • Most commonly used, due to better capillary action, is a multilayered metal mesh. • Sintered powder capillar structures allow the stronges capillar action, but are more expensive to manufacture. Even with this design, performance of the heat pipe still depends on the orientation, and upside-down operation usually still isn't possible.
Not only has the wick itself influenced the strength of the capillary action, but also the choice of working fluid. A high surface tension of the working fluid improves performance. Poor behaviour in case of overheating - if the heat source is too strong, and the heat pipe is heated beyond its specified temperature, it may happen that no condensation occurs anymore, and the entire working "fluid" inside the heat pipe is in vapour form. In this situation, the heat pipe's ability to transport heat collapses. This is a danger to consider when relying on heat pipes for cooling sensitive elements. Usage of heat pipes Given the excellent thermal conductivity of heat pipes, there are two situations in which heat pipes are particularly suitable: • In restricted space environments - in notebook computers, game consoles, home theater PCs, it is a common situation that due to space restrictions, it is not possible to mount a heatsink directly at the place where the heat is produced (that is, on the CPU). Here, a heat pipe can lead heat to a place where more space for a heatsink is available. • To improve thermal transfer within a heatsink: Heat pipes can be used in heatsink design to improve performance, by optimizing thermal conduction within the heatsink construction. This way, heat is more efficiently transported away from the contact area with the CPU, to the outer parts of the heatsink where cooling occurs. However, there have been PC coolers where it seems that the usage of a heat pipe was more of a marketing argument than a technical necessity. Heat pipes used in PC CPU coolers are usually inexpensive units with low capillary action. Considering this, the usage of a heat pipe on the CPU cooler, especially when installed in a tower case (vertical orientation of the heat pipe), is rather questionable.
Figure 4: Starting geometry for the problem. 1D Plug Flow First, this show sets up a 1D adiabatic plug-flow model describing the cup mixing temperature of the air (fluid), Tf,cup, in the channel between the boards during forced convection. It uses the equation.
T f ,cup = ( ∫ (( ρn.u.)T f dΩ 2 ))( ∫ ( ρn.u.)dΩ 2 ) −1 The model does not include the temperature distribution in the air. In addition, the model assumes the sources are infinite in the board’s lateral direction. Thus, in principle the model describes the distribution of temperature in the flow direction along a line in the air channel. Figure 5 depicts the resulting 1D geometry.
Figure 5: The geometry of the 1D-model. The model uses the General Heat Transfer application mode. It sets the convective velocity to 0.667 m/s at the inlet (that is, the average velocity of the previous models) and assume that it varies with temperature according to
u ( x) = u 0
T f , cup T0
The next equation describes the heat transfer
∇.(−k∇T f ,cup ) = Q − ρC p u.∇T f ,cup
Figure3. Unusual heat pipe-based cooler from Taisol: In this design, the fan is located between CPU and heatsink, and blows air away from the CPU, through the heatsink. Thermal transfer from CPU to heatsink is done by heat pipes.
Modelling: In the first stage, we have performed the simulation on without heat pipe based cooler. In brief, the system cools a stack of circuit boards with four in-line ICs, each producing 1W of heat, through forced convection. The aim of both of the following models is to determine the temperature development of the board and ICs.
Where k represents the thermal conductivity; Cp gives the specific heat capacity; and Q is the heating power per unit volume. The model sets Q to zero for the subdomains between the sources, and it equals 1666.67 W/m2 (that is, 2/3*1 W/(20e-3^2 m2)) at the source subdomains. The factor 2/3 represents the lateral average heating power, taking the open slots between the ICs into account. The material properties are the same as those in the previous paper models, also taking temperature variations into account. At the inlet boundary, the temperature is fixed to 300 K, and at the outlet the model applies convective heat flux. The goal is to calculate the ICs’ surface temperature, Ts. It is a function of the fluid temperature and the adiabatic heat transfer film coefficients, had, according to
Tc =
q had
+ T f ,cup
Where; q is the heat flux. This equation calculates the IC surface temperature. This model calculates values of had using the results of the 3D model with the formula:
Qtot A2 D (Ts − T f ,cup )
had =
Results and Discussion 1D plug flow Figure4 shows the results of the 1D model for the ICs’ surface temperature.
Where A2D is the IC’s xy-projected area and Qtot is the IC’s total heating power (in this case, 1 W). We can easily perform these calculations using the postprocessing capabilities of COMSOL Multiphysics. Specifically, with the discussed model, first calculate Tf,cup for each cross section of the air domain. Next calculate the total heat flux, Qtot, from each surface. Finally use the equation just given to derive the following values of had (a discussion of h0 follows shortly): Region
had [W/m2K]
ho [W/m2K]
Source 1
57.5
35.0
Source 2
44.4
22.1
Source 3
40.6
19.2
Source 4
39.2
18.0
Simplified 3D model This second model sets up a transient 3D model describing the temperature of the board and ICs during startup. In this case the modeled geometry consists of the board and ICs but not the air. The simplified model makes it possible to investigate the temperature transient of an entire row of ICs.
Figure 6: Geometry of the 3D model. In contrast to the model in the previous section, which used both conduction and convection, this model works only with the conduction feature of the General Heat Transfer application mode. It uses the isothermal film coefficients, h0, to calculate the convective cooling. We calculate them from the results of the 3D model, doing so in a similar way as we did for the 1D-plug-flow model just described except using the formula
h0 =
Figure 7: 1D model results for the surface temperature of the ICs (dashed line) and the average temperature of the fluid (solid line).
The profile agrees rather well with that of the previous 3D model, in this case experiencing a maximum surface temperature of 357 K. This indicates that you can model the heat transfer with good accuracy in a simplified way if you know the values of the film coefficient, had. The simplified 1D model is thus a good predictor even though it does not simulate the temperature distribution in the fluid and the fluid flow field. Simplified 3D model This model results in an accurate determination of the source surface temperatures. A benefit of having an easy-tosolve model is that you can proceed and analyse the transient behavior. Figure 8 shows the transient 3D model results at 1000 s.
Qtot A2 D (Ts − T0 )
Where; T0 is the air’s inlet temperature. To model the heat transfer coefficient of the board the built-in heat transfer coefficient library is used. A function valid for forced convection on plates is used. Further, the material properties specified in the subdomain settings for this model are identical to those in the previous models. The initial temperature of all components is 300 K, as is the surrounding temperature. For the ICs it applies a volume heat source of 1.25 MW/m. In the heat flux boundary conditions, for the downside segments of the board and for the circuit surface boundaries, it uses the h0 values.
Figure 8: Temperature of the source surfaces 1000 s after applying the heat load according to the simplified 3D model.
The results indicate that this amount of time is approximately sufficient to reach steady state.
Figure9. Temperature profiles at the surface of the extrusion without heat pipe in top and temperature profiles at the surface of the extrusion with heat pipes, in the bottom.
The temperature profiles at the surface of the extrusion without and with heat pipes appear in Figures 9a and 9b, respectively. The shape of the temperature profiles in Figure 9b demonstrates that heat pipe heat sinks can handle centered power input while maintaining a relatively uniform heat sink temperature. With heat pipe assisted heat sinks, the thermal designer has the freedom to place the electrical components in convenient locations and is not limited by spreading resistance to the center of the heat sink. As shown in Figure 9b, the heat pipes also significantly reduced the maximum surface temperature of the extrusion. By simply adding heat pipes to the base of an existing heat sink, the overall sink to ambient temperature rises are reduced by 30%. Faced with high spreading resistance, the thermal designer would probably explore the feasibility of increasing the base thickness in an effort to reduce both the heat transfer by conduction and by mass. Activity Migration Model Excessive junction temperature reduces reliability and can lead to catastrophic failure. We examine the use of activity migration which reduces peak junction temperature by moving computation between multiple replicated units. Using a thermal model that includes the temperature dependence of leakage power, we show that sustainable power dissipation can be increased by nearly a factor of two for a given junction temperature limit. Alternatively, peak die temperature can be reduced by 12.4 oC at the same clock frequency. The model
predicts that migration intervals of around 20–200 _s are required to achieve the maximum sustainable power increase. We evaluate several different forms of replication and migration policy control. In this section, we use the thermal model to evaluate the possible gains from activity migration. We only consider the case where we implement activity migration by pingponging between two sites. More sites could be used and would give further benefits but would increase area overhead. By pingponging between two sites, the activity for each site is halved, which means half the switching power density and half the temperature increase from the isothermal point. The same benefit could be achieved without pingponging if the computational load could be parallelized across the two sites. We are primarily interested in improving single thread performance and so do not consider parallel execution further. Reduced power density could be used either to increase performance or to reduce temperature. Figure 3 shows the qualitative benefit of activity migration in terms of new operating points. First, we can lower die temperature at the same clock frequency, which will also lead to lower leakage power. We can also exploit the time slack due to cooler operation by lowering VDD slightly to save some active power. Second, we can burn more power to increase clock frequency while maintaining lower temperature than the baseline. As an example of how this can be used, consider a laptop with limited heat removal capability. When running from battery power, we can use AM to lower die temperature which lowers leakage and allows more energy-efficient execution. When plugged into a wall socket, we can use activity migration to allow more power to be burned to raise performance without raising die temperature. There are many techniques that provide powerperformance tradeoffs such as dynamic voltage scaling (DVS), dynamic cache configuration modification, fetch/decode throttling, or speculation control [3]. We selected DVS as it is the simplest technique to illustrate the potential benefit of AM. To get the maximum benefit from the frequency-power tradeoff, supply voltage and threshold voltage should be scaled simultaneously. But for circuits with high activity factor, dynamic supply voltage scaling alone is adequate, and hence we assume fixed threshold voltages. • Activity Migration (AM) by turning on and off active power of hotspot and duplicated blocks (P_act1 and P_act2) • Identical thermal resistance and capacitance • Identical leakage power at same temperature: Figure shows the equivalent RC circuit for AM technique. Hot spot block 1 and duplicated block 2 have equivalent thermal resistances and capacitances. To determine the maximum achievable benefit, the duty cycle of pining is fixed at 50%. Bottom figure shows the benefits of AM analytically. AM drops temperature from Tbase. Temperature fluctuates between Thigh and Tlow. If period is small enough, Halve temperature will increase and so double sustainable power.
the thermal designer to utilize a recently accepted heat sink alternative, the heat pipe heat sink assembly. The heat pipe heat sink assembly permits the designer to achieve otherwise unreachable thermal resistance levels via the efficient maximization of the heat sink surface area available for dissipation to the ambient air.
Duplicated Block
HotSpot Block
(Tj1)
Active Power
(Tj2)
P_act1
Pam
Pbase
P_act2
0
Figure10. RC equivalent circuit
Time Increased sustainable power by AM + Perf-Pwr Tradeoff
AM Only Figure shows the equivalent RC circuit for AM technique.
Temperature
Tj1 Tj2
Active Power Pbase 0
P_act1 P_act2 Time
Temperature Tbase
Migration Period
Tiso
Time
Active Power
Reduced Temperature
Tj1 Tiso
Tbase
Tj2
Migration Period
P_act2 - long
Time
Hot spot block 1 and duplicated block 2 have equivalent thermal resistances and capacitances. To determine the maximum achievable benefit, the duty cycle of pining is fixed at 50%. Bottom figure shows the benefits of AM analytically. AM drops temperature from Tbase. Temperature fluctuates between Thigh and Tlow. Thigh equation is shown here and as you see if period is small enough, we halve temperature increase or we can double the sustainable power. Tbase − Tiso Period 1 + e 2τ
Time
Temperature Tbase
Temp can be reduced till (Tbase+Tiso)/2
Figure11. the equivalent RC circuit for AM technique.
Thigh =
P_act2 - short
Pbase 0
+ Tiso
−
Conclusion Heat pipes have an effective thermal conductivity many thousands of times that of copper. The heat transfer or transport capacity of a heat pipe is specified by its “Axial Power Rating (APC)’’. It is the energy moving axially along the pipe. The larger the heat pipe diameter, greater is the APR. Similarly, longer the heat pipe lesser is the APR. Heat pipes can be built in almost any size and shape. When used properly, heat pipes can do wonders. However, they are certainly not the ultimate solution to all cooling related problems. Due to the number of factors to consider when applying heat pipes, our advice is: Use ready-made heat pipe-based coolers only if you are absolutely sure that they are suitable for particular cooling problem. Heat pipes offer an attractive approach to supplementing conventional heat sink solutions. They provide a tool that allows the designer to reconfigure and/or extend the performance of more conventional heat sinks such as extrusions or castings. When performance dictates, they allow
Tj2 - long
Migration Period
Tiso
Time
Migration Period: AM + Perf-Pwr Tradeoff Active Power P_act2 - long
Pbase 0
Time Sustainable power can be increased till 2*Pbase
Temperature Tbase Tj2 - short Tj2 - long Migration Period
Tiso
Time
Figure12. Migration analysis
Historically, the application of this technology has implied significant costs; however, with the widespread application of the technology, costs are falling rapidly. Typical heat pipe heat sink assemblies are used in the portable computer market, where the highest production volumes currently exist. Higher performance units for workstation/server applications are several times that cost (for several times the performance).
The application of heat pipe heat sinks can in almost every instance yield improved heat sink efficiency and performance. This improvement, either in terms of permitting higher packaging densities or permitting higher power dissipation in a given volume, must be evaluated against the increased cost for such an assembly. We have examined the use of activity migration, which reduces peak junction temperature by moving computation between multiple replicated units. We show that sustainable power dissipation can be increased by nearly a factor of two for a given junction temperature limit. This increased power capacity can permit dynamic voltage scaling to achieve nearly 16% clock frequency increase using 180 nm technologies. The model predicts that migration intervals of around 20–200 _s are required to achieve the maximum sustainable power increase, and that this interval will scale down with future fabrication technologies. REFERENCES [1] M. Barcella et al. Architecture-level compact thermal R-C modeling. Technical Report CS-2002-20, Univ. of Virginia Comp. Sci. Dept., Jul. 2002. [2] Predictive technology model. Technical report, UC Berkeley, 2001. http://www-device.eecs.berkely.edu/ ptm/. [3] D. Brooks and M. Martonosi. Dynamic thermal management for high-performance microprocessors. In HPCA, Jan. 2001. [4] D. Burger and T. M. Austin. The SimpleScalar tool set, version 2.0. Technical Report CS-TR-97-1342, Univ. of Wisconsin Comp. Sci. Dept., Jun. 1997. [5] J. Deeney. Thermal modeling and measurement of large high-power silicon devices with asymmetric power distribution. In IMAPS, Sep. 2002. [6] B. A. Gieseke et al. A 600MHz superscalar RISC microprocessor with out-of-order execution. ISSCC, pages 176–177, Feb. 1997. [7] Intel. Intel to introduce new technologies to reduce power consumption of mpus, Aug. 2002. http://www.esionline.com.sg/news/view/default.asp?newId=10. [8] S. Sair and M. Charney. Memory behavior of the SPEC2000 benchmark suite. Technical Report RC 21852, IBM, Oct. 2000. [9] T. Sherwood, E. Perelman, G. Hamerly, and B. Calder. Automatically characterizing large scale program behavior. In ASPLOS, Oct. 2002. [10] K. Skadron et al. Control-theoretic techniques and thermal-RC modeling for accurate and localized dynamic thermal management. In HPCA, Feb. 2002. [11] Standard Performance Evaluation Corp. CPU2000, 2000. [12] C. Weaver. SPEC2000 Alpha binaries (little endian). http://www.eecs.umich.edu/˜chriswea/benchmarks/spec20 00.html [13] COMSOL software, version 3.1 [14] Kourosh Mousavi Takami, Evaluation of oil in over 20 year’s old oil immersed power transformer, Mazandaran University, May 2001.
[15]Kourosh Mousavi Takami, Advanced Transformer Monitoring & Diagnostic Systems and thermal assessment with robust software's, research presentation, Water and power University, March 2007, Tehran, Iran [16] Kourosh Mousavi Takami, Hot Spot identification and find a best thermal model for large scale power transformers, April 2006, KTH University, Stockholm, Sweden. [17]Kourosh Mousavi Takami, Hassan Gholinejad, Jafar Mahmoudi, Thermal and hot spot evaluations on oil immersed power Transformers by FEMLAB and MATLAB software’s, IEEE Conference, Int. Conf. 2007, London, 17 April 2007. [18]Kourosh Mousavi Takami, Jafar Mahmoudi, Evaluation of Large Power Transformer Losses for green house gas and final cost reductions, 3rd IGEC conference, Sweden, June 18, 2007. [19]Kourosh Mousavi Takami, Jafar Mahmoudi, A novel device (oil spraying system) for local cooling of hot spot and high temperature areas in power transformers, 3rd IGEC conference, Sweden, June 19, 2007. [20]Kourosh Mousavi Takami, Jafar Mahmoudi, Thermal evaluation and energy saving with loss reduction in core and winding of power transformers, 3rd IGEC conference, Sweden, June 19, 2007. Kourosh Mousavi Takami was born in Sari, Mazandaran,Iran . He received the B.S.c. degree in electric power engineering from the Iran University of Science and Technology (IUST) Tehran, Iran, Oct1995 and the M.Sc. degree in electric power engineering from the Engineering Faculty of Mazandaran University, Iran in 2002. Currently, he is PhD student at Mälardalen University in Sweden since 2005. He has so over ten years experience in power system design and installations.His research interests include Optimization and simulation of heat generation and transfer in the core and winding of power transformers; diagnostic testing and condition monitoring of power equipments, and application of fuzzy and Ants algorithm to condition monitoring of power equipments. Jafar Mahmoudi was born in Tehran, Iran. He received the B.Sc., M.Sc. Degree in Sharif University and PhD degrees from KTH University, Stockholm, Sweden. Currently, he is a Professor with the Department of Public Technology Engineering in MdH University, Västerås, and Sweden. His major research focus is development of new technology and methods for industrial energy optimization with special focus on heat and mass transfer. He has years of theoretical & experimentalexperience on this. He also has a broad technical background encompassing thermodynamic, numerical methods and modelling (CFD computation) as well as materials science. This in combination with his industrial experience has served as a solid basis to build upon in expanding his research activities and focusing on relevant and current industrial issues. Over the last 10 years his focus has been on the practical and industrial application of the above mentioned methods, an effort conducted in a large number of industrial projects. In this, his teaching experience has proved invaluable.
