576 TIIERMAL DESIGN CONSIDERATIONS
`
`• Ambient
`
`Board
`
`- . . Conduction
`
`~ Convection
`Radiation
`
`Figure 12-3 Heat transfer in a low performance multichip module.
`
`potentially multiple heat sources in thermal communication with each other via
`the substrate and its ambient
`The heat generated at the chips seeks the most conductive path to reach the
`sinks. The sinks to which the heat is eventually transferred are the cooling fluid
`and the boards. The paths available for heat flow are through the substrate then
`the molding material and the leads. The flow of heat is impeded by each
`material, regardless of its thickness, as it travels from the sources to the sink.
`As the heat reaches the leads, part of it is conducted to the board, and the rest
`is either radiated or convected to the ambient.
`The flow of heat spreads, seeking the greatest ratio of A to L by Equation
`12-2 to maximize the conductive heat transfer rate. As a side effect of this heat
`spreading, each chip has a higher temperature due to the presence of its
`neighbors. The closer the chips are spaced, the greater will be this effect.
`However, heat spreading also can be beneficial because of the increase in
`heat transfer rate. In plastic SCMs, this is done by inserting an aluminum plate,
`called a "heat spreader," into the package body. In some MCMs this is done
`with copper plates. As long as these copper plates conduct much more heat to
`the heatsinks than to each other, the net result is beneficial.
`A similar process occurs as the heat reaches the physical boundaries of the
`component Figure 12-3 gives a schematic depiction of different modes of heat
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 601
`
`

`
`THERMAL PHENOMENA IN ELECTRONIC ENCLOSURES 577
`
`transfer and heat flow process in an MCM. Combination of multiple heat
`sources and different possible avenues for heat flow have created a rather
`complex and nonuniform temperature field.
`
`12.3.3 The Concept of Thermal Resistance
`
`The concept of thermal resistance is associated with impeding the flow of heat
`through a medium. Analogous to electrical resistance, if resistance is decreased,
`less voltage is required to pass the current through the wire. Current is similar
`to heat flow, and voltage to temperature. Hence, if thermal resistance is
`decreased, smaller temperature differences across a medium result. Thus, it
`becomes intuitive that if thermal resistance is minimized, internally and
`externally, the junction temperature is reduced.
`The package thermal response can be viewed by two resistances, external
`and internal. The internal resistance, 0 je Gunction to case resistance), addresses
`heat transfer within the package, from the chip to the surface of the package.
`
`The external resistance, ° ea (case to ambient resistance), is a measure of thermal
`
`transport occurring between the package surface and the ambient.
`0 je and 0 ea are defined by the following equations,
`
`and
`
`0 je
`
`0 ea
`
`T
`J
`
`- Te
`P
`
`Te
`
`- Ta
`P
`
`(12-5)
`
`(12-6)
`
`In Equations 12-5 and 12-6, subscripts a, c and j refer to ambient, case and
`junction, respectively. The denominator, P, is the total power dissipation in the
`component. The units of these resistances are DCIW. Figure 12-4 shows a
`schematic representation of these resistances.
`There are three problems, however, with this use of the concept of thermal
`resistance. First, it is difficult to calculate a single number for thermal resistance
`with the presence of multiple heat paths. Second, even if you did calculate it,
`you could not reuse the resistance in other designs. Consider what would happen
`to the heat flow pictured in Figure 12-4 if another MCM were placed on the
`bottom of the board. Less heat would flow in this direction due to the smaller
`ilT. The total heat flow would change and the thermal resistance would increase.
`These problems are common to MCMs and SCMs [13]. In fact, a common error
`in SCM design is to assume that the data sheet value for thermal resistance
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 602
`
`

`
`578 TIIERMAL DESIGN CONSIDERATIONS
`
`Ambient
`
`Case to Ambient
`
`Junction to Case
`
`Figure 12-4 Thermal resistance representation in an electronic component.
`
`applies to a crowded two-sided board. It often bas to be increased by 50% or
`more to compensate for the ideal conditions under wbich the manufacturer
`measured it The third problem is unique with MCMs. With multiple chips,
`eacb with a different power, P, each chip will bave a different temperature, Tj,
`and thus the choice of values to be used in Equations 12-5 and 12-6 are arbitrary
`and somewbat meaningless.
`Thermal resistance is a widely used concept. The preceding discussion bas
`sbown the weaknesses associated with using Equations 12-5 and 12-6 even for
`SCMs. Tbese equations contain even higher error levels wben applied to low
`performance MCMs. In the case of bigh performance MCMs, thermal resistance
`in this form is completely useless. Although thermal resistances for these MCMs
`still are reported, it typically is for the cbip and not the entire module.
`Nevertheless, thermal resistance is a very valuable concept for qualitatively
`understanding the thermal effects of different materials and cooling approacbes.
`Equations 12-2 and 12-3 show conduction and convection beat transfer.
`Conductive and convective thermal resistances are defined as
`
`(12-7)
`
`(12-8)
`
`Rk
`
`=
`
`Rh
`
`=
`
`L
`kA
`
`1
`bA
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 603
`
`

`
`THERMAL PHENOMENA IN ELECTRONIC ENCLOSURES 579
`
`which refer to internal and external resistances. By defmition, Equations 12-7
`and 12-8 are identical to equations 12-5 and 12-6. We see that Rx is inversely
`proportional to k and A (area normal to the direction of heat floW). ~ is
`inversely proportional to h and A (convective surface area). To reduce these
`resistances, the denominator would have to increase.
`Consider an MCM where the chips are epoxied to the substrate. The
`thermal resistance between the chip and the substrate (where the heat is
`conducted away) is the one imposed by the epoxy. The area, A, associated with
`the epoxy is constrained to the size of the chip. Often times in the application
`of epoxy, air gaps are created adding to overall chip to substrate resistance. The
`resistance is reduced by removing the air gaps or using an epoxy with a higher
`thermal conductivity or another bonding technique. The Hitachi Silicon Carbide
`(SiC) RAM [14] is an example of such practice where 52 solder bumps are used
`for heat transfer purposes only (Figure 12-5). Of course, as we change process
`or epoxy, we have to be concerned with material compatibility to avoid uneven
`expansion. Otherwise, stresses induced as the result of uneven expansion may
`
`Heat Sink
`
`1
`
`SiC
`Silicone
`Ceramic
`Layer
`/
`\
`r.:~~~~~~~~~~~;-Lead Frame
`
`•
`
`Silicon Circuit
`Board
`
`Memory
`Chip
`
`Figure 12-5 Cross sectional view of Hitachi air-cooled multichip module.
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 604
`
`

`
`580 THERMAL DESIGN CONSIDERATIONS
`
`result in component failure. For this reason, thicker epoxies typically are used
`in laminate MCMs.
`Similarly, if we look at Rh, it can also be reduced by increasing h or A.
`The heat transfer coefficient, h, is increased by going to higher velocity flows
`(fans or jet impingement) or changing the fluid (gas to liquid). The surface area
`is increased by adding heatsinks to the component.
`Based on the above discussion, thermal resistance plays a pivotal role in the
`magnitude of junction temperature. Reducing thermal resistance either internal
`or external to the MCM package impacts thermal control positively.
`In the
`design of MCMs, it is important to locate the heat sources and the thermal
`resistances on their paths. The resistances should be reduced so that minimum
`spreading takes place, and the path from the chip to the sink has the least
`thermal resistance.
`
`12.3.4 Heat Transfer On a Board
`
`For thermal design purposes, each component cannot be considered in isolation.
`The heat being produced by one component is transferred amongst the others.
`This is referred to as thermal coupling and is discussed here at the board level,
`and in the next section at the system level. As discussed above, thermal
`coupling can be very strong within an MeM.
`Consider the forced air cooled board shown in Figure 12-6. There are two
`mechanisms leading to thermal coupling. First, the MCMs and SCMs heat the
`air as it passes over them. The downstream parts will experience a hotter fluid
`
`Figure 12-6 Flow over a circuit board containing SCMs and MCMs.
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 605
`
`

`
`THERMAL PHENOMENA IN ELECTRONIC ENCLOSURES 581
`
`temperature, Tf and, according to Equation 12-3, the component is hotter, by the
`same amount, to dissipate the generated heat. Second, some of the heat
`produced by each component is passed to the board through leads, conduction
`and radiation across the air gap. Though part of this heat is convected away
`from the board, part of it is conducted to the other components on the board,
`raising their temperature. While the board dielectric is a poor conductor of heat,
`the copper layers within it are excellent heat conductors. An eight layer board
`exhibits a stronger conductive thermal coupling between components than does
`a four layer board.
`The net effect of this is that the chip with the highest Tj might not be the
`chip that produces the most heat. The critical chip is the one that has a Tj
`closest to or over the specified limit. For example in Figure 12-6, the high
`power CPU core MCM is placed near the fan while the low power main
`memories are placed at the other end. Often this arrangement requires that
`special memories be purchased with an aluminum heat spreader within them.
`
`12.3.5 Thermal Coupling in Electronic Enclosures
`
`To appreciate the impact of the system (enclosure) on the thermal performance
`of MCMs, it is necessary to review the thermal phenomenon in an enclosure.
`(See Figure 12-7 for an example of an enclosure.)
`The shelf or card holder (cage) is where a circuit board resides in the
`system. Boards are normally inserted into the shelves through card guides.
`Except in some specialized cases where a latching mechanism is used to rigidly
`attach the board to the shelf, the boards are loosely fitted inside the shelf.
`Therefore, the necessary contact to facilitate conduction heat transfer from the
`board to the shelf does not usually exist.
`The backplane, or motherboard, in a PC is another avenue for the heat to be
`transported to the ambient or the shelf. If the thermal conductivity of the board
`is very large, that is multilayered boards with several layers of copper,
`conduction heat transfer through the backplane can be significant. However, the
`thermal coupling by convection and radiation heat transfers is significantly larger
`than conduction heat transfer.
`Frames or enclosures that house single or multiple shelves generally are
`designed to be isolated from the shelves. Thus, the heat generated within the
`system normally is convected through the vent holes. Although this constitutes
`the bulk of heat flow, there exists significant thermal coupling between the
`boards (and shelves) and the frame. The thermal coupling, in the order of
`significance, is by radiation, convection and conduction heat transfer. Since the
`frame is in contact with the system ambient, it acts as a sink and source of heat
`for the system.
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 606
`
`

`
`582 THERMAL DESIGN CONSIDERATIONS
`
`Inlet Grill
`
`Power Supply
`
`Figure 12-7 Schematic view of electronic system configuration.
`
`The magnitude of conduction heat transfer is very system dependent. The
`radiation heat transfer, however, is generally the predominant mode of thermal
`coupling between the shelves and the frame. The radiation heat transfer tends
`to be even more significant if the system is cooled by natural convection.
`The frame is coupled to the surrounding ambient via radiation and
`convection heat transfers. The system ambient also can act as a source and a
`sink. The magnitude of these heat transfers varies significantly with the changes
`in the system surroundings.
`The thermal transport process in electronic systems is quite involved and can
`become very complex. Because of many different thermal processes and strong
`coupling at various system levels, thermal bookkeeping is necessary for accurate
`design. In addition, it should be clear that we cannot only focus on a component
`(module) without considering the system, environment and other parameters
`affecting thermal design.
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 607
`
`

`
`THERMAL MANAGEMENT OF MCMS 583
`
`12.4 THERMAL MANAGEMENT OF MCMs
`
`Power dissipation levels exceeding 4 W/cm2 and specific system performance
`requirements have forced design of highly customized cooling systems for some
`MCMs. MCMs typically have higher power dissipations than SCMs and
`generally are placed on circuit boards or substrates that contain other potentially
`high power components. The propensity for thermal spreading within the circuit
`board or the substrate has sometimes led the designers to build individual cooling
`elements for every chip on the substrate. This combination at times has created
`a challenging problem for thermal management of MCMs. The challenge has
`embraced packaging of the MCM itself and integration of its cooling system into
`the overall frame.
`In low performance systems, thermal management of the MCM is typically
`an after thought and is constrained by the application. In high performance
`systems, the cooling system design is an integral part of the design cycle.
`Because of the customized nature of cooling systems, their spatial restrictions in
`terms of system compactness or electrical distance between high speed
`components has been an added challenge to thermal management.
`The constraints imposed by temperature limits and system compactness have
`produced innovative packaging and thermal control methods. In this section,
`some of these techniques, with emphasis on cooling method, are presented The
`objectives of thermal management, with respect to design and manufacturing
`constraints, are discussed. Then, the cooling approaches and order of application
`are reviewed.
`
`12.4.1 Alternate Thermal Control Methods for MCMs
`
`There are alternate thermal control methods for both the internal and external
`paths. The choice of the primary internal path is closely related to the choice of
`chip attach. The primary internal path alternatives are (Figure 12-8) through the
`substrate, through the substrate with thermal vias or thermal cutouts, and chip
`backside.
`With through the substrate cooling, the main heat dissipation surface in
`contact with the fluid is beneath the chip. The primary internal heat path is
`through the substrate. There might be considerable heat spreading and the
`thermal resistance might be high, particularly as the thermal conductivity of the
`most common substrate materials is low. It can be reduced by using different
`materials, such as aluminum nitride instead of alumina. However, highly
`thermally conductive alternatives to polyimide and laminate MCM materials are
`not available.
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 608
`
`

`
`584 TIIERMAL DESIGN CONSIDERATIONS
`
`Through-the-substrate:
`
`--
`
`Chip Backside:
`
`II!!
`
`Figure 12-8 Through-the-substrate and chip backside technology for MCMs.
`
`If the material properties cannot be improved, then the heat spreading and
`the thermal resistance can be reduced, at the expense of wiring capacity, by
`placing copper vias in the substrate or by sinking the chip into the substrate.
`Either of these techniques can be used with any chip connection technique. If
`wire bond or TAB leads are used, the chip can be attached with a thermal epoxy.
`Part of the heat is transferred through the thermal epoxy and part through the
`metal leads. With solder bump attachment, the heat is transferred through the
`bumps and the air gap. Often extra bumps are added to reduce thermal
`resistance.
`With chip backside cooling, the main heat dissipation surface is above the
`chip and is attached with a metal mount, a high thermal conductivity solder, a
`thermal epoxy or a thermal grease. Only flip techniques (flip TAB or solder
`bumps) can be used because of potential damage to the surface of the chips.
`Chip backside cooling generally has the smallest internal thermal resistance.
`In either thermal path, careful attention must be given to the interfaces. For
`example, if the epoxy chip interface has many air bubbles in it, the thermal
`resistance increases substantially.
`In general, the external thermal control methods can be categorized as
`follows:
`
`• Natural convection
`•
`Forced convection
`• Conduction or radiation cooling
`•
`Liquid immersion
`•
`Phase change (boiling)
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 609
`
`

`
`Table 12·2 Thermal Control Methods.
`
`THERMAL MANAGEMENT OF MCMS 585
`
`Primary Cooling
`Mechanism
`
`Typical
`HTC
`(W/m2K)
`
`Relative
`Effectiveness
`
`Achievable
`Density
`
`Complexity
`
`Natural convection
`(air)
`Forced convection
`(air)
`Natural convection
`(liquid)
`Forced convection
`(liquid)
`Phase change
`(liquid)
`
`10
`
`100
`
`100
`
`1000
`
`5000
`
`0.1
`
`1.0
`
`1.0
`
`10.0
`
`50.0
`
`Low
`
`Very low
`
`Medium
`
`Low
`
`Medium
`
`Medium
`
`High
`
`High
`
`High
`
`High
`
`Natural convection cooling is the case when no fluid movers are used in
`circulating the fluid in the system. Cooling by forced convection utilizes a fluid
`mover to circulate the fluid. Conduction or radiation cooling is when a cold
`plate [15] or radiation plate is used to remove heat. The application of the latter
`is seen in military and space electronics. Liquid immersion is when a component
`or system is immersed in a liquid. The liquid can be fluoroinerts or others such
`as liquid nitrogen. With boiling, the fluid boils at the MCM contact surface.
`General features of the convection-based cooling modes are given in Table 12-2
`[16]. In that Table, HTC is the heat transfer coefficient h (see Equation 12-3).
`A designer often is confronted with the decision of selecting a cooling
`method, keeping in mind manufacturing issues and end use application
`constraints. Selecting a thermal control method is a function of component
`temperature rating and the heat removal capacity of a specific design. The
`coolant fluid, gas or liquid, typically sets the capability of these cooling methods
`apart. This is made more clear by looking at Figure 12-9 [17]. The figure
`merges the thermal control methods with power dissipation (heat flux) and
`temperature rise. It shows a representation of the expected temperature rise over
`ambient for different cooling methods. It also hints to a potential thermal control
`method as a function of component power density.
`Figure 12-9 does not suggest or provide an absolute case for a thermal
`It suggests the expected range or type necessary for heat
`control method.
`removal to ensure the junction temperature meets its constraint. For example,
`for an MCM with 10 W/cm2 and a temperature rating, Tj , of 85°C, Figure 12-9
`suggests some sort of liquid cooling. However, the same component can be
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 610
`
`

`
`586 TIIERMAL DESIGN CONSIDERATIONS
`
`Q)
`
`u 0
`u c
`....
`
`Q)
`
`~ -~
`::I -0 ....
`
`Q)
`
`....
`
`Q)
`Q.
`E
`Q)
`I-
`
`8
`
`4
`
`2
`
`2
`
`4
`
`e 8 _I
`10
`
`2
`
`• e I
`
`2
`
`10
`
`Surface heat flux t W/CM 2
`
`Figure 12-9 Temperature differences attainable as a function of heat flux for various heat
`transfer modes and various coolant fluids.
`
`effectively cooled with a high level air system Get impingement. where the air
`is blown directly onto each heatsink [18]) and may not require exotic cooling.
`The high end air-based thermal control methods tend to be not as expensive as
`liquid based ones, yet they yield junction temperatures within the rated limits.
`This has sparked much interest in the community to explore and expand air
`cooling limits.
`To provide an overall view of cooling techniques practiced in the industry,
`it is worthwhile to review some of the designs and bighlight their salient thermal
`management features. References [19] and [20] provide an excellent overview
`of this subject and excerpts from these and other references are used in the
`forthcoming discussion. Figure 12-5 shows Hitachi's SiC RAM representing low
`performing systems. The module bas six 1 W chips that provide 1 kbit of
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 611
`
`

`
`THERMAL MANAGEMENT OF MCMS 587
`
`memory. The cooling system is an 8 mm high 4 fin simple heatsink with air as
`the coolant fluid. Using a finned heatsink increases the area in contact with the
`coolant and thus reduces Tj. The chips contain 77 solder bumps; 52 of them are
`there purely for thermal reasons. The bumps act as thermal paths, carrying the
`dissipated heat from the chips to the substrate. The substrate's thermal
`conductivity is approximately 14 times that of alumina, thus it is very effective
`for spreading the heat [20]-[21].
`
`Cap
`
`Heat Sink
`Thermal
`Conductive
`Adhesive
`
`i
`
`\
`
`Module
`Substrate
`
`Chip
`
`Low Melting
`Point Solder
`Copper Plate
`
`Figure 12-10 Cross sectional view of Mitsubishi air-cooled high thermal conduction
`module.
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 612
`
`

`
`588 THERMAL DESIGN CONSIDERATIONS
`
`Figure 12-10 shows another example of air cooled modules designed by
`Mitsubishi known as the High Thermal Conduction Module. The component's
`total power dissipation is 36 W generated uniformly in nine 3K gate ECL chips.
`The internal thermal paths consist of both the top and bottom of the chips. The
`chips are placed on bumps to accommodate heat transfer from the bottom side
`to the substrate. The combination of a copper plate heat spreader placed on the
`top of the chips (Figure 12-10), and a cap heatsink assembly provide thermal
`paths from the top side of the chips. The heat generated by the chips is
`dissipated through the heatsink and convected away by air with a velocity of 6
`mls.
`References [19] and [22] discuss the details of heat transfer analysis and
`subsequent thermal resistances for this module. A note worthy point that relates
`to our earlier discussion is the variations in heat flow paths and thermal
`resistance. Reference [22] reports that in the absence of heat transfer from the
`pins to the air, the central chip conducts 13% of its heat through the solder
`bumps versus 18% for the peripheral chips. The central chip has 3°CIW and
`peripheral chips have 2.5°CIW chip to heatsink thermal resistance, resulting in
`O.5°CIW difference in thermal resistance for these chips. The combination of
`internal multiple heat paths (stemming from heat spreading) and the nature of the
`air flow through the heatsink accounts for this difference. The analysis presented
`in the noted references has ignored the heat path to the PCB via the module's
`pins.
`Inclusion of this heat path may have led to an even higher difference
`between these chips.
`Although the chip power dissipation is uniform, the boundaries of the
`peripheral chips are different from the central one. The central chip is
`surrounded, on all sides, by 4 W chips. The peripheral ones have at least one
`side facing the periphery of the module. Therefore, as a result of thermal
`coupling through the substrate and the surrounding gas, the central chip is
`expected to operate at a higher temperature. As a side note, this example typifies
`the difficulty of reporting a single thermal resistance value for MCMs.
`The IBM 4381 module, Figure 12-11 [19], is an example of a high
`performance air cooled MCM that uses higher level air cooling (impingement).
`The module is dimensioned 64 mm x 64 mm x 40 mm and contains 36 chip
`sites. The chips are solder bumped on a multilayered ceramic (MLC) substrate
`and separated from the ceramic cap by a layer of thermal paste designed by
`IBM. The power dissipation per module varies from 36 W - 90 W resulting in
`a typical circuit board power dissipation of 1.3 kW. Reliable system operation
`requires that the chip temperature not exceed 90°C.
`The chips are closely spaced to minimize delay. The power dissipation per
`chip is high (3.8 W maximum). For this power dissipation level and temperature
`rating constraint, each chip requires individual thermal management and control.
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 613
`
`

`
`THERMAL MANAGEMENT OF MCMS 589
`
`Impinging Air Flow
`
`Thermal ~~~~~~?lt
`Grease
`
`Chip(s)
`
`Ceramic
`Cap
`MLC
`Substrate
`
`Figure 12-11 Impingement air-cooled MCM used in mM 4381 processor.
`
`The heatsink, air impingement combination provides an adequate level of thermal
`control at the chip level to ensure chip temperature below 90°C.
`Perhaps the most talked about example of liquid cooling is the IBM Thermal
`Conduction Module (TCM), used in the 3081 processor, Figure 12-12 [19]. The
`design requirement specified an 85°C temperature limit for the chip and achieved
`69°C. The system performance required that up to nine TCMs be mounted on
`a single PCB. Since each TCM can dissipate up to 300 W, the total power on
`the PCB is 2700 W. Stringent chip temperature limit and high system
`performance was the driving force behind development of this MCM. The
`thermal design objective was reached by removing heat from the chips as directly
`as possible and minimizing thermal spreading. The heat dissipated by each chip
`is conducted via the spring loaded piston in a helium atmosphere to the water(cid:173)
`cooled heat exchanger, Figure 12-12.
`Fujitsu's FACOM M-780 is a water cooled MCM that departs markedly
`from the IBM and other similar water cooled modules (Figure 12-13) [19] and
`[23]. In this design, the thermal control unit consists of bellows and water jets
`packaged in a closed system. The tip of the bellows is in contact with chip
`surface through a compliant material to ensure adequate thermal contact.
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 614
`
`

`
`590 TIIERMAL DESIGN CONSIDERATIONS
`
`Figure 12-12 IBM thermal conduction module (TCM) with water cooled cold plate.
`
`The FACOM M-780 has 336 single chip modules mounted on both sides of
`540 mm x 488 mm PCB. The maximum chip power is 9.5 W and the board
`dissipates 3,000 W [19]. The cold plate is introduced to the section of the PCB
`containing the single chip modules. The cold plate is factory assembled and
`cannot be separated for field repairs.
`The next level of cooling that a few computer companies have gravitated
`toward is liquid immersion. The same criteria drive the selection of the thermal
`control method: system performance and temperature limit. One of the
`advantages of immersion cooling is eliminating interface resistances seen in the
`cold plate thermal control methods. By immersing the MCM or the entire circuit
`board in a Fluorocarbon (pc-n, FC-77), immersion cooling is attained. The
`most noted forced immersion cooled system is the CRAY-2 supercomputer.
`The immersion cooled portion of the CRA Y -2 consists of SCMs mounted
`on eight PCBs, dissipating a total of 600 - 700 W for a heat density of 0.21
`W/cm2• Though this is within air cooling limits, the large air flow rate required
`would have been impractical to design. Hence the CRA Y -2 uses FC-77
`fluorocarbon cooling forced horizontally over the surface of the PCBs with 2.5
`cm/sec. velocity.
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 615
`
`

`
`THERMAL MANAGEMENT OF MCMS 591
`
`ChiP(S~~II~II§~Lead(S)
`
`Heat
`Transfer
`Plate
`
`Printed
`Circuit
`Board
`Compliant
`Spacer
`
`Figure 12-13 Cross sectional view of Fujitsu water-cooled bellows cold plate cooling
`system.
`
`Before closing this section, it is important to revisit a few points regarding
`the thermal control techniques used for MCMs. We looked at many examples
`from the simple application of a flat finned heatsink to immersion cooling of
`MCMs. Two issues should be evident by now. First, thermal control techniques,
`beyond the use of a heatsink, are system dependent, or customized. This
`dependency stems from system level packaging and performance requirements.
`For example, CRAY-2 designers determined that with 0.21 W!cm2 heat flux, air
`cooling was possible. Volumetric air flow requirements, however, were not
`practical and a liquid cooling system was designed. Therefore, customization of
`thermal control methods suggests that it is not safe to defme general design rules.
`The second point is thermal resistance minimization. In all the examples
`reviewed here, and in many more, the designers have gone through much effort
`to ensure that the internal and external resistances are as small as possible. An
`example to be cited is the IBM TCM, in which helium is used in place of air to
`improve conduction heat transfer through the gas. By replacing air with helium
`inside the TCM, the internal thermal resistance was reduced from 25°C/W to
`8.08°C/W [24]. The importance of this point cannot be emphasized enough.
`Use of different chip mounting technologies and materials (discussed in other
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 616
`
`

`
`592 lHERMAL DESIGN CONSIDERATIONS
`
`chapters) for internal cbip design can have a major impact in thermal
`performance of the MCM. Hence, examination of alternatives, with regard to
`thermal tradeoffs, should be a routine exercise for an MCM designer.
`
`12.4.2 Cooling Methods - Cost Impact of
`Thermal Management Techniques
`
`The electronics industry can be segmented into four categories:
`
`• Computer
`• Military and Space
`• Telecommunications
`• Consumer products
`
`MCMs have been and will continue to be used in products produced by these
`industries. The thermal control technique is a function of system application and,
`therefore, varies between each industry. The high end computer industry has led
`the way in cooling system design and can afford to use exotic, although not
`desirable, cooling methods. The military has close tolerance requirements,
`resulting in unique and system specific cooling systems. Telecommunications
`tends to gravitate toward lower level cooling methods. Consumer electronics
`seeks passive cooling techniques because of their application. With the increase
`in processing speed, all these industries seek bigher order cooling methods. This
`trend will continue until significant changes in the packaging of electronics
`components are introduced.
`The power density, system packaging and junction temperature specifications
`set the foundation for selection of a cooling method. Additionally, this selection
`is constrained by manufacturability and cost. The hierarchy of the cooling
`methods are:
`
`• Air in natural convection, with or without heatsink
`• Air in forced convection, with or without heatsink and other flow
`enhancement methods
`• Air jet impingement
`• Radiation
`• Liquid cooling:
`Natural convection
`Forced convection
`Jet impingement
`Immersion
`Immersion by boiling
`•
`• Cryogenics
`
`msgalica@mintz.com
`
`Elm Exhibit 2162, Page 617
`
`

`
`THERMAL MANAGEMENT OF MCMS 593
`
`The above list is ordered in the ease of implementation, but in a reverse order
`of cooling capacity (see Figure 12-9). Natural convection, with air as the
`working fluid, is most desirable since no cooling system design is required. This
`does not imply that thermal analysis is not needed. It suggests that no cooling
`system (fluid movers, heat exchanger etc.) is required for thermal management
`of the system. Forced air convection is the next most desirable mode. Use of
`heatsinks with air cooling can further increase the heat removal capacity of
`forced air convection. Some industry segments (telecommunications and
`consumer products) tend to shy away from fans (fluid movers) because of
`reliability issues and noise. Fluid movers are unavoidable if power dissipation
`and temperature specifications are at such a level that fans are required.
`From jet impingement to higher cooling methods, heat removal capability
`goes up significantly. But implementation in a system from cost, manufacturing
`and user impression becomes complex. For example, jet impingement requires
`a compressor and placement of jet nozzles throughout the circuit board, creating
`a reliability and physical design dilemma. Additionally, there is a whistling
`noise as the air expands and leaves the nozzle. Imagine your PC or workstation
`whistling continuously - fan noise is uncomfortable enough.
`Liquid cooling is a very attractive proposition for high powered components
`and systems. But impleme