Patent Application
20050127500
This application contains subject matter which is related to the subject matter of the following co-pending application, which is assigned to the same assignee as this application, International Business Machines Corporation of Armonk, N.Y. An application entitled “Improvement in chip cooling” having attorney docket number YOR920020329US1 filed on Dec. 27, 2002, is hereby incorporated herein by reference in its entirety.
This invention relates to cooling within integrated circuit (IC) packaging structures. More particularly, the present invention is directed to the cooling of integrated circuit chips using a relatively thick compliant thermally conductive material where there is a nonuniform power distribution on the integrated circuit.
As heat is generated during the functioning of integrated circuit chips (ICs), the thermal resistance to the heat sink must be small so that the operating temperature of the chip is low enough to assure the continued reliable operation of the device. The problem of heat removal becomes more difficult as chip geometries are scaled down and operating speeds are increased, resulting in increased power density. The ability to adequately cool the chips is therefore a limiting factor in the further increase of system performance. Integrated circuit chips mounted on substrates, and particularly in an array on a substrate such as is found in a multi-chip module (MCM), present special cooling difficulties. In an MCM, the chips may be mounted very close together and nearly cover the whole top surface of the MCM. With such an arrangement, it may not be possible to use a heat spreader bonded directly to the back surface of the chips, as is sometimes used for isolated chips, to reduce the heat flux (power/unit area, i.e. W/cm2). An additional problem is that the processor, and other chips, frequently have a “hot spot” which can have a heat flux significantly greater than the average heat flux resulting in temperatures of about 20° C. hotter than the average chip temperature. A thermal solution which maybe adequate for the average chip power density may not be adequate to allow reliable operation of the hot spot region of the chip.
A common technique for removing heat from high-power IC's makes use of a cooling plate or heat sink which is thermally attached to the chips using a compliant thermally conductive material. Heat is removed from the cooling plate or heat sink by methods such as forced air cooling or circulating liquid coolants. Due to the difference in thermal expansion between the chip and the material of the cooling plate or heat sink, a compliant thermally conductive material is required therebetween. This is especially true for MCMs where an array of chips is mounted on a common substrate and power cycling can result in both vertical and horizontal deflections of the cooling plate relative to the back surface of the IC's. For a chip in a single chip module (SCM) type package, a thermal spreader composed of a high thermal conductivity material with a thermal expansion coefficient close to that of Si, such as SiC, is sometimes bonded to the chip using a silver (Ag) filled epoxy or other thermally conductive adhesive. Due to the close thermal expansion match, a rigid bond with an adhesive can be used, as a compliant layer therebetween is not needed. A heat sink is then mounted onto the thermal spreader using a layer of a compliant thermally conductive material. As an alternative for single chip modules, where mechanical deflections are minimized and the package structure keeps the two surfaces very parallel to each other, a very thin layer, such as 25 microns or less, of a compliant thermally conductive material can be used.
Compliant thermally conductive materials are typically thermal pastes, or thermal greases and are frequently referred to as thermal interface materials (TIMs). Thermally conductive pastes typically comprise thermally conductive particles having a distribution of sizes dispersed within a binder material or matrix, such as the paste described in U.S. Pat. No. 5,098,609 issued Mar. 24, 1992 to Iruvanti et al. In the '609 patent, paste is applied between the top of the IC mounted on the substrate and the lower flat surface of a cooling plate facing the substrate. Typical TIMs include those having a wax matrix, commonly known as phase-change materials, those having a silicone-based matrix, and dry particle lubricants such as graphite and metal powders.
When applying a compliant thermally conductive material between the back of a chip which is electrically attached to a substrate and the lower surface of a cooling plate or heat sink, the compliant thermally conductive material layer is preferably made as thin as possible in order to reduce the thermal resistance through the layer. The compliant thermally conductive material is preferably flexible, maintains its integrity, surface adhesion and chip coverage despite the expansion and contractions of the packaging structure caused by power and temperature cycling. A difficulty observed with the use of compliant thermally conductive materials is the migration of the material from behind the chip and the formation of voids due to differential thermal expansion of the various parts of the package during power and temperature cycling. Such migration of the compliant thermally conductive material can greatly increase the thermal resistance between the chip and the cooling plate or heat sink during the lifetime of the electronic package, possibly causing catastrophic heating and failure of the chip. This migration of the compliant thermally conductive material limits the minimum thickness of material being required to insure reliable operation, with thinner gaps resulting in earlier failures. This critical thickness is a function of the amount of expansion and contraction present in the package, the chip size, the properties of the compliant thermally conductive material, the expected operating life of the package, along with other factors.
In U.S. Pat. No. 5,825,087 issued Oct. 20, 1998 to Iruvanti et al. and assigned to the present assignee, the cooling plate used in conjunction with a thermal paste or thermal adhesive has been roughened by grit blasting or provided with a plurality of crisscrossing channels in order to improve the adhesion of the thermal medium and inhibit its flow during operation of the electronic module.
In U.S. Pat. No. 5,247,426 issued Sep. 21, 1993 to Hamburgen and Fitch, an apparatus comprising a non-uniform thermal conductance structure which includes high thermal conductance regions and low thermal conductance regions is disclosed where the apparatus is coupled to a semiconductor to establish a desirable temperature profile across the surface of the semiconductor.
U.S. Pat. No. 5,623,394 issued Apr. 22, 1997 to Sherif et al. and assigned to the present assignee, is directed at customizing the cooling of different chips on a MCM by using a plurality of thermally conductive materials. U.S. Pat. No. 5,757,620 issued May 26, 1998 to Edwards et al. and assigned to the present assignee, is also directed at customizing the cooling of different chips on a MCM by varying the depth of thermal compound filled gaps, or blind hole that is above each chip. Both of these do not address the problem of needing a reduced thermal resistance at the “hot-spot” of a chip but are directed at adjusting the thermal resistance only at the chip level.
In U.S. Pat. No. 5,668,404 issued Sep. 16, 1997 to Abe and Ohmae, a semiconductor device includes a semiconductor chip attached to a lead frame with recesses on the rear surface of the semiconductor chip opposite the lead frame. These recesses increase the heat radiation area of the semiconductor chip.
U.S. Pat. No. 6,255,695 issued May 1, 2001 to Chia et al. relates to a flip-chip semiconductor package where the non-active surface of the semiconductor die has a plurality of grooves formed thereon; and a heat sink attached to the non-active surface of the semiconductor die with an adhesive. The grooves increase the contact area of the adhesive to the chip, thereby increasing the mechanical bond strength and thermal conductivity between the semiconductor die and the heat sink.
A publication entitled “Using cap-integral standoffs to reduce chip hot-spot temperatures in electronic packages”, 2002 Inter Society Conference on Thermal Phenomena, IEEE 2002, pp. 173-178, by M. S. June & K. K. Sikka, who are both employed by the present assignee, and the subject matter of which issued as U.S. Pat. No. 6,294,408 entitled “Method for Controlling Thermal Interface Gap Distance” and assigned to the present assignee, presents a technique of reducing the chip hot-spot temperatures using cap integral standoffs. For an advanced thermal paste interface in a high power electronic module dissipating 100 W, standoffs can reduce the hot-spot temperature by 5-10° C. The standoffs are essentially columns of a high thermal conductivity material in parallel with a relatively low conductivity material. Each standoff will either directly contact the chip or will have an extremely thin interface between itself and the chip. This is due to the low cross sectional area of the standoff which tends to displace any paste that gets under the standoff. The disadvantages of this approach include the precision machining required on the heat sink to add the standoffs and the requirement to precisely align the standoffs with the desired locations on the chip.
Thus, from the prior art discussed above, it is desired to reduce or eliminate “hot spots” associated with active areas of processing chips, while maintaining a compliant interface between the chip and a substrate (e.g., cold plate, heat sink, or other cooling mechanism) in thermal communication therewith, and to prevent the formation of voids in the compliant thermally conductive material during assembly or subsequent thermal or power cycling.
The shortcomings of the prior art are overcome and additional advantages are provided through the provision of forming raised portions on the back surface of an integrated circuit in a pattern which corresponds to the hotter areas on the chip. This results in a projecting “mesa” structure on the back side of the chip so that after module assembly, the thickness of the compliant thermally conductive material layer is locally thinner over the “hot spots”. This results in a local reduction of the thermal resistance and hence a reduced temperature of the hot spot on the chip. By local reduction of the paste layer thickness, the allowable global paste layer thickness can be increased, or maintained at a thicker level, which reduces the likelihood of migration of the compliant thermally conductive material and the formation of voids between the chip and the heat sink or cold plate and insures that adequate mechanical compliance is provided between the chip and the heat sink or cold plate. Additionally, if individual pistons are used (as described in U.S. Pat. No. 6,214,647) to better control the compliant thermally conductive material layer thickness over each chip, additional small raised portions at the corners or along the edges of the chip can be formed to insure that the compliant thermally conductive material layer is uniform and that the piston, or heat sink, or cold plate, is not tilted relative to the chip surface.
While the preferred embodiment references using mesas over high heat flux areas to reduce the maximum temperature of such areas, typically found in the core areas of processor chips, this same technology may be used to locally lower the temperature of specific functional areas of the chip that have serious repercussions if that function fails even though the area may have just average heat flux densities. For instance, a chip function that is non-redundant or difficult for the server to recover from if it fails could have its circuits cooled lower by placing a mesa over this functional area, improving the reliability of these sensitive circuits.
Additional features and advantages are realized through the techniques of the present invention. Other embodiments and aspects of the invention are described in detail herein and are considered a part of the claimed invention. For a better understanding of the invention with advantages and features, refer to the description and to the drawings.
The subject matter which is regarded as the invention is particularly pointed out and distinctly claimed in the claims at the conclusion of the specification. The foregoing and other objects, features, and advantages of the invention are apparent from the following detailed description taken in conjunction with the accompanying drawings in which:
The detailed description explains the preferred embodiments of the invention, together with advantages and features, by way of example with reference to the drawings.
The present invention is directed at locally reducing the thickness of a compliant thermally conductive material in a region or regions aligned with high power density areas on a chip while avoiding migration of the material from behind the chip and the formation of voids.
As transistors are scaled down in size and as the operating frequency increases, the power density (W/cm2) of processor chips will continue to increase. For reliable long-term operation, it is necessary to remove the heat produced by a chip and keep the junction temperatures below about 105° C. Note that the acceptable junction temperature is a function of the technology used and the reliability requirements of the specific product.
Referring now to FIGS. 4(a) and 4(b), a semiconductor package structure 100 is illustrated. Structure 100 includes a chip 112 that is in thermal communication with a substrate 114 on one surface, while an opposite surface is operably connected to a C4 array 116 for connection to a module (not shown). In an exemplary embodiment, substrate 114 is a heat sink, cold plate, or other suitable cooling means, or a lid, or heat pipe, or other intermediate structure which is in contact with a suitable cooling means. The main thermal dissipation path from chip 112 includes a thermal interface material (TIM) or thermal paste layer 118 which provides mechanical compliance and provides stress relief, between chip 112 and substrate or thermal cap/heat sink 114. The type of paste described in U.S. Pat. No. 5,098,609 can be used in the present invention, as can other compliant thermally conductive pastes used in the art or other compliant conformable thermally conductive materials. Furthermore, it is contemplated that TIM 118 includes a fluid when chip 112 is at operating temperature, such that phase change materials are contemplated.
For the case when the back surface 119 of chip 112 is substantially flat having no raised portions, the thermal paste layer 118, between the back surface 119 of the chip 112 and the structure 114, is about 70 percent of the internal thermal resistance, Rint, if no thermal spreader is used and 114 is a lid or thermal hat. Rint represents the internal thermal resistance of structure 100: the resistance through the chip 112, through the thermal interface material 118, and to the top surface of the lid structure 114, to which a cooling means is attached. When a thermal heat spreader 120 is attached with a thermally conductive adhesive, such as, but not limited to, for example, an Ag filled epoxy layer 122 to the back of the chip 112 as illustrated in
Referring now to
The simplest implementation is to pattern the back surface 119 of a processor, or other, chip 112 while it is in wafer form as that is a batch process using available techniques and avoids the alignment issues present with patterning the thermal cap/heat sink 114, and additionally, fine features such as microstructure can be incorporated onto the mesa region as will be described further. Given that the desired feature sizes are larger than about 2-5 microns, a contact printer (not shown) can be used for backside lithographic processing on the wafer, which is less expensive than step and repeat exposure tools. In one exemplary embodiment, photolithography is used to pattern a masking layer such as silicon nitride on the back side of the wafer using reactive ion etching. The wafer is than mounted in a fixture which provides a liquid tight seal around the back edge of the wafer and a wet etch, such as TMAH (tetramethyl ammonium hydroxide) and water is used to etch the Si. Other suitable semiconductor substrate material/etchant combinations may also be used. The masking layer could be then removed using either a wet etch and the same fixture, or if desired, a dry etch step could be used. The depth of recess being contemplated is about 25 to about 75 microns (the current paste thickness is about 100+/−25 microns. The use of an anisotropic Si wet etch such as TMAH results in a tapered sidewall (at an angle 136 of about 55 degrees to the back surface 119 of chip 112) and facilitates the formation of self-terminated microstructures on the top surfaces of the mesas. Alternately, isotropic Si wet etches or Si dry etching is optionally used to form the mesa's, or the mesa's could be formed by a additive process where material such as a solder or metal is added to the back surface of the wafer discussed more fully hereinbelow.
In an exemplary embodiment with reference to
In an alternative embodiment shown schematically in
In a further alternative embodiment shown schematically in
A series of experiments was performed using single chip modules (SCM) and thermal test chips which were about 18×18 mm in size and had temperature monitors centered in each quadrant of the chip and were equipped with heaters to provide a uniform heat flux in each chip quadrant. Chips were fabricated with two mesas, each about 50 microns tall, in diagonally opposite chip quadrants where the size of the mesa was varied, either 4×4 or 6×6 mm, and the mesas were either solid (no microstructure) or were patterned with microstructure. The microstructure consisted of either what is referred to as “groove” microstructure which actually are a regular array of 10×10 micron pyramidal pits on 20 micron centers etched into the top surface of the mesa, or what is referred to as “half tone” microstructure which are actually a checkerboard pattern of 100×100 micron truncated pyramidal pits etched into the top surface of the mesa. By checkerboard, each 100×100 micron truncated (50 micron deep) pyramidal pits is surrounded on four sides by unetched regions and, with a slight separation, additional truncated pyramidal pits are diagonally adjacent. The parts were assembled with a nominal thermal paste layer thickness of 37.5 microns over the surface of the mesas and the corner support posts using thermal caps as described in U.S. Pat. No. 6,214,647. During the assembly and encapsulation process, the single chip modules were thermal cycled to cure a polymer sealant which attached the cap to the ceramic substrate on which the thermal chips were mounted. The thermal resistance measurements found that the average thermal resistance over a mesa was reduced by 22% compared to a site where there was no mesa. When the mesa had either the “grooved” or “half toned” microstructure, the average thermal resistance was reduced by 32% compared to the sites with no mesa. It should be noted that the measured thermal resistance includes not only the thermal resistance across the compliant thermal paste layer, but also includes the thermal resistance through the silicon chip and through the copper (Cu) lid structure. It is contemplated that the improved performance for the mesa structures with microstructure is perhaps due to reduced void formation during the assembly and encapsulation process. Alternatively, it is contemplated that the increased Si surface area in contact with the complaint thermal paste results in a reduction in the thermal interface resistance. As described previously, this is unlikely given the large bulk thermal resistance of the thermal paste compared to the bulk Si thermal resistance.
In further alternative embodiments, the raised mesa region(s) 132, 142, 152, 162 on the back (non-active) surface 119 of the chip 112 can be formed by an additive process such as by locally depositing, or depositing and patterning, metallic or other highly conductive layers. This could be done by electroplating in a masking layer of a dry film resist, for example.
If small gap technology is used as it is disclosed in U.S. Pat. No. 6,214,647, which is incorporated by reference herein in its entirety, it may be preferable to not form any structure on the piston which is used to adjust the gap as this could result in the piston tilting or could require the piston to be have a fixed rotation to correctly align which would reduce the ability of the piston to adjust for any tilt or height variation of the chip in setting the thermal paste gap. If structures are formed on a surface in contact with the piston, it maybe desirable to form small raised portions 142 at the corners of the chip 112 so as to insure that the paste gap 144 is uniform and the piston is not tilted relative to the surface which it is contacting.
None of the prior art known teaches locally reducing the thermal resistance as a function of location within a chip by structuring the back of the chip for a compliant thermal joint, but instead attempts to reduce the overall thermal resistance. The above described disclosure describes an apparatus and method for reducing the peak temperature on a chip with a compliant thermal joint. In an exemplary embodiment, the temperature of the “hot spot” is reduced by locally reducing the compliant paste layer thickness aligned therewith. The above described disclosure further details an apparatus and method to counter a critical thickness or a limit of how thin a compliant thermal interface material (TIM) layer disposed between the chip and substrate may be with a MCM. This critical thickness is a function of the amount of horizontal and vertical expansion/contraction present, chip size, and the TIM properties. The use of a raised portion or “mesa” as disclosed above, especially with microstructure on the mesa, allows a local reduction of the paste thickness without failure due to paste pumping if the area of the mesa is less than or equal to about 25 percent of the total chip surface area and not located directly along a chip edge, so that a “dam” of normal thickness paste is formed before the free edge to reduce pumping. More specifically, an edge defining each raised portion 132 and interfaced with back surface 119 of chip 112 is at least 1 mm away from an edge defining chip 112.
While the preferred embodiment to the invention has been described, it will be understood that those skilled in the art, both now and in the future, may make various improvements and enhancements which fall within the scope of the claims which follow. These claims should be construed to maintain the proper protection for the invention first described.
