Patent Application
20100304565
The present invention relates to semiconductors and, more particularly, to electrical connections for such devices.
Making electrical contacts that extend all the way through an electronic chip (by creating electrically conductive vias) is difficult. Doing so with precision or controlled repeatability, let alone in volume is nearly impossible unless one or more of the following is the case: a) the vias are very shallow, i.e. significantly less than 100 microns in depth, b) the via width is large, or c) the vias are separated by large distances, i.e. many times the via width. The difficulty is compounded when the vias are close enough for signal cross-talk to occur, or if the chip through which the via passes has a charge, because the conductor in the via can not be allowed act as a short, nor can it carry a charge different from the charge of the pertinent portion of the chip. In addition, conventional processes, to the extent they exist, are unsuitable for use with formed integrated circuit (IC) chips (i.e. containing active semiconductor devices) and increase cost because those processes can damage the chips and thereby reduce the ultimate yield. Adding further to the above difficulties is the need to be concerned with capacitance and resistance problems when the material the via passes through has a charge or when the frequencies of the signals to be carried through the vias are very high, for example, in excess of about 0.3 GHz.
Indeed, there are numerous problems that are extant in the semiconductor art including: use of large, non-scaleable packaging; assembly costs don't scale like semiconductors; chip cost is proportional to area, and the highest performance processes are the most expensive, but only fraction of chip area actually requires high-performance processes; current processes are limited in voltage and other technologies; chip designers are limited to one process and one material for design; large, high power pad drivers are needed for chip-to-chip (through package) connections; even small changes or correction of trivial design errors require fabrication of one or more new masks for a whole new chip; making whole new chips requires millions of dollars in mask costs alone; individual chips are difficult and complicated to test and combinations of chips are even more difficult to test prior to complete packaging.
Accordingly, there is a significant need in the art for technology that can address one or more of the above problems.
We have developed a process that facilitates joining of chips to each other.
One aspect involves an apparatus. The apparatus includes a semiconductor wafer that has been back-end processed, the semiconductor wafer including a substrate, electronic devices and multiple metalization layers, a via extending from an outer surface of the substrate through the substrate to a metalization layer from among the multiple metalization layers, and an electrically conductive material within the via, the electrically conductive material forming an electrically conductive path from the metalization layer to the outer surface.
Another aspect involves a method of processing a semiconductor wafer that has been front-end and back-end processed. The method involves forming a via in the semiconductor wafer extending from a surface of the wafer, into and through semiconductor material, to a metalization layer formed during the back-end processing by etching the semiconductor wafer; and making the via electrically conductive so as to form an electrical path within the via extending from the surface of the wafer to the metalization layer.
The advantages and features described herein are a few of the many advantages and features available from representative embodiments and are presented only to assist in understanding the invention. It should be understood that they are not to be considered limitations on the invention as defined by the claims, or limitations on equivalents to the claims. For instance, some of these advantages are mutually contradictory, in that they cannot be simultaneously present in a single embodiment. Similarly, some advantages are applicable to one aspect of the invention, and inapplicable to others. Thus, this summary of features and advantages should not be considered dispositive in determining equivalence. Additional features and advantages of the invention will become apparent in the following description, from the drawings, and from the claims.
At the outset, it is to be understood that the term “wafer” as used herein is intended to interchangeably encompass all of the terms “chip”, “die” and “wafer” unless the specific statement is clearly and exclusively only referring to an entire wafer from which chips can be diced, for example, in references to an 8 inch or 12 inch wafer, chip or die “-to-wafer”, “wafer-to-wafer”, or “wafer scale” processing. If use of the term would, as a technical matter, make sense if replaced by the term “chip” or “die”, those terms are also intended. Moreover, a substantive reference to “wafer or chip” or “wafer or die” herein should be considered an inadvertent redundancy unless the above is satisfied.
In general, specific implementations of aspects described herein make it possible to form connections among two or more wafers containing fully-formed electronic, active optical or electro-optical devices in a simple, controllable fashion which also allows for a deep via depth, high repeatability, controlled capacitance and resistance, and electrical isolation between the via and the wafer or substrate through which the via passes.
Implementations of our process make it possible to form an electrically conductive via that is narrow in width (i.e. down to about 15 microns wide or less) as well as deep (i.e. to more than about 50 microns in depth) through a chip of depth to width ratios on the order of 3:1 and as much as 30:1, although aspect rations on the order of 5:1 to 10:1 will be more typical. Moreover, our approach advantageously makes it possible to do so in circumstances where the portion of the chip the via passes through will be electrically active. Specifically, we make it possible to provide electrical access through the doped semiconductor part of a wafer using a passage where side-walls insulate the doped semiconductor from an electrical conductor which propagates through the passage. Moreover, our process works for narrow passages (i.e. about 15 microns wide or, in some cases less) while allowing for tight control of the thickness of the isolating material and the electrical conductor so as to maintain a constant and acceptable capacitance and resistance.
Still further, our approach is suitable for use in forming contacts having, if circular, a diameter of between 0.1 micron to 15 micron pads, the upper end not being a limit but rather simply the size below which our approach permits integration not generally possible with other approaches, and the lower end being a function of currently available photolithography technology. In other words, advances in photolithographic technology that allow for smaller definition will also allow the current limit to go smaller.
Still further, and unlike solder contacts, which can be hundreds or thousands of microns long, or wirebond contacts, which can also be thousands of microns long and thus often require significant pad drivers to drive the impedance between chips, through our approaches, we can use very short contacts (10 microns or less) which allows much lower parasitic electrical effects between the chips. Our typical contact has spacing between contacts three times or less the width of the malleable material (defined and discussed below) prior to integration with a complementary contact (e.g. if the initial contact is 8 microns high, spacing between contacts would be up to about 25 microns.
Our approach further permits stacking of chips on a separation spacing of less than or equal to about 20 microns. In practice, less than or equal to 10 micron spacing will be typical, although we have demonstrated that less than about 1 micron spacing can be done. In general, the minimum is determined by the topology of the closest surfaces of the two wafers being joined; when they are touching at their highest points the distance between the pads represents the maximum height spacing.
Our approach further makes it possible to form contacts on a pitch of less than or equal to 50 microns. Typically, pitches of less that or equal to about 25 microns will be used, although we have demonstrated that pitches as small as 7 microns can be done, again that limit being a function of currently available photolithography technology. Here too, as technology advances, pitches can be smaller.
Features of some variants include one or more of the following: potential for millions of contacts/cm2; electrical, mechanical and thermal attachment occurs concurrently; attachment done with low force but yields high strength connection (on the order of 1,000 kg/cm2); connections can be done with economies of scale; non-planar wafers can be accommodated; most processing can be done on a wafer scale (e.g. 10 micron GaAs on 8″, 10″ or 12″ wafers); processes can be done on a chip to chip, chip to wafer, or wafer to wafer basis; processes are electrically grounded; connections are made on a pre-formed (i.e. device bearing chip) so can be used with third-party supplied chips; making of vias before multiple chips are connected; capability to test chip combination before it is permanently connected and to rework if necessary; mixing and matching of different technologies (i.e. GaAs to InP, InP to Si, GaAs to Si, SiGe to SiGe to Si, etc. and even an insulator wafer made of, for example, ceramic, LCP or glass); an ability to create chip-sized packages that take advantage of semiconductor process economies; ability to allow low-speed functions to be moved off of core, expensive processes, but still have entire set of circuits act like a single chip, allows design of an individual chip to take advantage of the variety of voltages, technologies, and materials available and best suited for that particular design; irrespective of the technologies required for other aspects of the design; enhanced off-chip communication; facilitates increased modularity of design at the chip level allowing leverage of core designs into multiple products without having to absorb redundant non-recurring engineering costs; and allows matching of speed with technology type so that low speed circuitry need not be formed on expensive, higher speed technology than necessary.
In overview, our processes improve the ability to create a chip-to-chip connection using “through-wafer” electrical contact that can be used with a doped substrate but will not short out the substrate and thus can carry an opposite charge to that of the substrate through which it passes. In addition, this “through-wafer” approach is usable with wafers of semiconductor materials, insulators such as ceramics, and other conductive or non-conductive materials. Moreover, using current equipment for etching semiconductor materials, i.e. having a 30 to 1 aspect ratio, the process works well for vias of narrow cross section (i.e. 15 microns wide, or in some cases less) and vias extending for an overall depth from in excess of 50 microns to depths of 500 microns or more. In addition, the process allows for close control of capacitance and resistance such that, for example, the vias created using the process can carry high speed electrical signals (i.e. of frequencies in excess of 0.3 GHz) or, in some implementations, optical signals.
Some implementations will also allow for concentric vias that, if conductive, can each carry different signals or different charges. Still further, some implementations allow for concentric vias in which the inner via can be used to as part of a cooling system by using a part of the arrangement to become part of a heat pipe arrangement. Other implementations provide the advantage that they are compatible with, and allow use of, stacking approaches in which chips are stacked and electrically connected to other chips on a chip-to-chip, chip-to-wafer or wafer-to-wafer basis.
Advantageously, virtually all of the stacking processes and variants described herein, or straightforwardly derived therefrom, only require a new stacked piece to be aligned to the piece directly below it. This is in sharp contrast to prior art techniques that attempted to stack and which must align all pieces in the stack together and then insert a conductive material to form the trans-stack connections. Such an approach requires all of the pieces in the stack to be accurately aligned with respect to every other piece in common rather than just to the piece below it. Moreover, our approaches work equally well with uniaxial, coaxial and triaxial connections, whereas alignment in-common approaches do not, if they can be done at all.
The various approaches are described for simplicity by way of example, using examples involving wafers of semiconductor material, for example silicon (Si), silicon-germanium (SiGe), gallium-arsenide (GaAs), etc., that have been pre-formed (i.e. they already contain integrated circuits or their components, and/or optical devices such as lasers, detectors, modulators as well as contact pads for those devices).
The first example of the approach involves a two-etch process where only wafer, for purposes of example semiconductor material (i.e. doped semiconductor with or without some or all of its associated substrate), needs to be etched. This example process begins with a device-bearing wafer of semiconductor material. One or more trench regions of precise width are etched in the wafer to the desired depth such that, in the case of a semiconductor wafer, the trench extends into the wafer substrate and creates a perimeter about a portion of the semiconductor material. Notably, the shape of the perimeter can be any closed shape and the outer and inner walls of the trench need not be the same shape. Capacitance and resistance of the ultimate via connection can be controlled through selection of the shape of the inner and outer perimeter of the trench and their separation distance(s). The trench depth is typically 50 microns or more, in some cases 500 microns or more, but the trench does not propagate through the entire substrate of the wafer so that the bounded semiconductor piece doesn't fall out. The trench is then filled with an electrically insulating material. At least a portion of the bounded semiconductor piece is then etched away leaving a hole of narrower cross section than that bounded by the outer trench wall, such that the via created by etching the semiconductor piece is bounded either by insulating material or a perimeter ring of material from the center semiconductor piece for part of its depth and substrate for the rest. The hole is metalized to create an electrical connection between the top of the wafer and the bottom of the hole. The back of the wafer (i.e. the substrate) is then thinned to expose metalization at the bottom of the hole which then becomes a substrate side contact or a portion thereof (interchangeably referred to herein by the broad term “contact”). Typically, at least the full depth of a portion of the surface defining the hole will be metalized, although in some implementations the metalization will only extend to a sufficient depth that it will be exposed when the substrate is sufficiently thinned. In this manner, if the process used to perform the metalization can not be used to metalize down to the full depth, as long as sufficient metalization extends down to where the thinning will stop, the contact can be formed. For example, in one example implementation, if the via extends partway into the substrate for a total length of about 600 microns, but the metalization can only be reliably done to an overall depth of about 300 microns (i.e. 300 microns less than the via itself), the process is not adversely affected so long as the substrate can be thinned to at least reach the metallization without unacceptably weakening the wafer or chip.
Through the above approach, variants described herein, and permutations and combinations thereof, connection points can be brought closer to the on-chip devices. By bringing connection points closer to on-chip devices, this approach facilitates chip-to-chip connections in the vertical direction (i.e. through chip stacking), can reduce the distance between connection points, and reduce or eliminate the need to use wirebonds for chip to chip connections. Moreover, the approach facilitates creation of sub-component specialty designs that can be mixed and matched as desired during production. In other words, a third dimension becomes more readily available for chipset materials, geometries and manufacture. In addition, the approach enables mixing of different speed or types of material technologies as well as mix-and-matching of component or subcomponent designs thereby providing development and manufacturing cost savings. Still further chip-to-chip connections can be created that use optical rather than electrical connections between chips.
The above is further facilitated through the optional use of a chip-to-chip connection approach that reduces the stress on chips being joined, thereby reducing the risk of chip damage.
The particular aspects described above are illustrated in greater detail by way of a number of examples and with specific reference to figures which, for purposes of illustration and clarity of presentation, are overly simplified and not to scale. In some cases, the scales are intentionally grossly exaggerated or distorted at the expense of accuracy for enhanced clarity of presentation and understanding.
Moreover, the approaches described herein are independent of the particular devices on the chip or with which the aspects described herein are used. Thus, the references to any specific type of device, for example the laser of the first example, are arbitrary and irrelevant to the aspects described herein except to the extent that they are devices to which electrical contact may need to be made. In other words, the approaches described herein are essentially identical for all devices and circuit elements to which contact may be made.
As shown, the laser 104 is a conventional vertical cavity surface emitting laser (VCSEL). For purposes of explanation, it should be assumed that the top mirror 106 will need to be electrically connected to some element on the side 118 of the substrate opposite the side 120 carrying the laser 104 and pass through the doped semiconductor material 122 near the device 104 within a specified area 124.
At the outset, it should be understood that to the extend lasers or photodetectors are discussed as the devices, the terms “top” and “bottom” follow a convention whereby the “bottom” is the portion closest to the substrate, irrespective of whether the laser emits towards or away from the substrate 112 (or in the case of a photodetector the direction from which it receives light).
The basic process of forming the through-chip contact will be described with reference to those aspects introduced in
First, a trench 302 is etched into and through the semiconductor material 122, preferably using an anisotropic etching process (in order to create relatively straight trench sidewalls 304), to a depth that brings the trench 302 part way into the substrate 112. The overall depth of the trench 302 can be 100 microns or more, in some cases extending for 500 to 600 microns or more. However, the trench 302 should stop before extending completely through the substrate 112 otherwise the ability to implement the invention can, in many cases, be lost. The trench 302 is shaped such that it is closed on itself creating a cross section in a plane parallel to the plane of the substrate that is an annulus. Through use of this annular trench 302, an “island” 306 of the semiconductor material 122 will remain and be held in place at least by the intact part 308 of the substrate 112. At this point it is worth noting that, while the “annulus” referred to for the trench 302 is shown as circular in shape, this is only for purposes of simplicity of illustration. As used herein, the terms “annular” or “annulus” should be understood to not be limited to any particular or regular shape nor does the outer periphery have to have the same shape as the inner periphery. As long as the trench is a closed shape so that it creates an isolated “island” within it, the trench is to be considered an annulus trench or “annular” as used herein. In other words, the terms are intended to include any combination of closed perimeter shapes including closed polygons (regular or irregular) or other closed perimeter shapes whether, for example, the shape is smooth, erose, etc. Moreover, the terms are intended to encompass fixed and varying widths as needed or desired for the particular instance.
At least the trench 302 is coated with a dielectric or other electrically insulating material 500, which can optionally also cover a portion of the top outer surface 116 to a desired thickness. Optionally, if heat transfer is a concern a material that, while electrically insulating, is a good thermal conductor may be used as the electrically insulating material 500.
Advantages achieved by the above approach can be appreciated when viewed in contrast in the context of the prior art. First, as a general matter, it is extremely difficult to apply dielectric materials in a uniform manner, particularly where a uniform thickness is required. Second, this problem is compounded when the dielectric needs to be applied to a non-flat surface and is further compounded when they must be applied to vertical walls, such as those of the vias described herein. Thus, to the extent other approaches attempt to create holes and then accurately coat the walls of those holes with dielectric and thereafter make them conductive, those approaches lack any meaningful ability to control uniformity. The lack of uniformity present in those approaches dramatically affects capacitance and impedance, and hence performance, particularly where the signal frequencies involved will be very high, for example, in excess of about 0.3 GHz. In contrast, with the approaches described herein, precise control of capacitance and resistance is possible because the dimensions of the trench 302 can be precisely controlled to the precision of the trench 302 itself. The peripheral walls of the trench 302 define the thickness and uniformity in coverage of the insulating material 500 (and hence, the ultimate capacitance and impedance) because they constrain it. Therefore, all that is required is ensuring that the trench 302 is filled—a very low precision and low cost process. Thus, unlike the prior art, precision during application of the dielectric is unnecessary.
Once the electrically insulating material 500 has solidified (by hardening, curing or other processing), a via trench 702 is created by removing the island 306 of semiconductor material within the annulus 704 of insulating material 500 to a sufficient depth 502 necessary to achieve the particular desired implementation, for purposes of example, a depth similar in depth to that of the trench 302 (i.e. such that it too extends some distance into the substrate 112 but preferably not fully through it). In practice, the depth 502 of the via trench 702 can be longer or shorter than depth of the trench 302 provided it too extends sufficiently deep that it can be reached, if necessary during processing as described below, in this example case, essentially the same distance into the substrate 112 as the trench 302. Moreover, the innermost wall of the annulus 704 that bounds the island 306 dictates the shape and profile of the via trench 702 that is created by the removal process will be a dielectric. Accordingly, it will not typically be impacted by an etch process, a low precision etch process can be used to remove the island 306 of semiconductor material because rigorous control of the removal is unnecessary in the width or depth directions. Of course, removal can be augmented, or alternatively otherwise be accomplished, by using one or more other suitable processes, for example, laser ablation, laser drilling or some combination thereof.
Continuing with the process of this example, once the via trench 702 has been created, the sidewall(s) 706 of the via trench 702, as well as the bottom 708 of the via trench 702, will all be electrically non-conducting because the sidewall(s) 706 will be the insulating material 500 and the bottom 708 will be defined by the substrate 112.
The via trench 702 is made electrically conductive by “metalizing” at least a longitudinal portion of the via trench sidewall surface 706 (i.e. along its depth), for example, using sputtering, evaporation, plating or other physical or chemical deposition techniques for applying metals or, if need be, some combination thereof. In other words, the metalizing can involve use of a conductive solid, a conductive epoxy or a reflowable material (e.g. an appropriate temperature conductive liquidus like a solder). This metalizing process can, and typically will, be used to create a continuous electrically conductive connection from at least about the via bottom 708 to the upper surface 116, and in many cases, all the way to the device of interest if it is part of the chip in which the via was made. By way of representative example
As noted above, because the width and length of the insulating annulus can be rigorously controlled, as can the thickness of the conductor formed by the metalizing, a constant capacitance relative to the metalized surface can be achieved. Moreover, the insulating material 500 electrically isolates the contact 904 from the semiconductor material 122 it is passing through and thus, can account for defects in the semiconductor material that might otherwise electrically short the contact to another device or conductor.
There are several advantages that can be obtained through use of one or both of these optional approaches. First, filling the void with a material adds mechanical strength and increases structural rigidity thereby reducing potential stresses. Second, the use of solder, an epoxy or other bonding material can aid in the ultimate connection of the chip to another element, particularly when the connection involves hybridization of that chip to another chip. Third, by inserting a material into the void, the risk of undesirable materials entering the void is reduced. Finally, the filler material reduces or eliminates the possibility of damaging the metalized portion within the via trench, particularly if less than the total sidewall is metalized. In addition, by varying the thickness of the insulator and metal, the coefficient of thermal expansion (“CTE”) of the wafer can be balanced so as to match that of the wafer. For example, an oxide (CTE of 1 ppm) can be used in conjunction with copper (CTE of 17 ppm) to match the CTE of silicon (CTE of 2.5 ppm).
Of course, since these aspects are both optional, both can be dispensed with while still using the invention. For completeness of understanding however, both processes are illustrated in connection with
Once the metalization is complete, if the remaining void 1100 is not going to be left empty for use as described later, the remaining void 1100 can optionally be partially or wholly filled with some material, for example, in this case a bonding substance 1102. Depending upon the particular implementation this variant will be used for, the bonding substance 1102 can be conductive or non-conductive, i.e. an electrically conductive substance such as solder, metal or alloy that can be applied through, for example, electroless or electroplating techniques or deposited by evaporative deposition or sputtering, or a non-conductive bonding agent like, for example, an appropriate type of glue or epoxy or oxide like silicon dioxide.
Alternatively or additionally, if the metalization has not completely filled the void, once the metalization is finished, the remaining void 1100, if any, can optionally be partially or wholly filled with, for example, a simple finishing substance 1302. Depending upon the particular implementation this variant will be used for, the finishing substance 1302 can be, for example, an insulator such as the insulating material 500 that was initially used to fill the trench 302 a conductor such as a conductive epoxy, a conductive solid, or a reflowable material, otherwise a conformal coating can be used. In addition, the finishing substance 1302, if used, need not be introduced solely into the void 1100. As shown in
Returning to the basic process,
Once the metalization aspect shown in
Alternatively, the thinning can be performed until the bottom metallization 1502 is removed or the void 1100 volume is exposed (whether filed or not).
Alternatively, the arrangements of
It should now be appreciated that above basic process, as well as the more complex alternative processes that follow and build upon the basic process, provide a further advantage over the prior art in that making of the vias before fabrication of the devices (e.g. transistors, diodes, lasers, photodetectors, etc.) on the wafer is not required. Moreover, the process does not require that the vias only occur in on the periphery of the chip in areas where conventional wirebond pads would occur. Instead, the instant process is more localized and can be performed at sufficiently low temperatures such that circuitry can be formed on or embedded in the semiconductor before via formation and the vias can be placed in areas other than the periphery of the chip. This makes it possible to use the process with chips made by others, without the need to be involved in the design process of those chips, and as will be described in greater detail below, to make connection paths between devices on different chips much shorter than could be done through the use of wirebond pads. Still further, because the process facilitates making paths through the wafer, as described in greater detail below, the process is highly useful for chip stacking or for creating mix and match chip “units”.
One problem that can arise in connection with the filling of a trench with an electrically insulating material, particularly when the trench is narrow in width and relatively deep, for example 100 microns or more in depth, is the possibility of there being pinholes, air bubbles or other imperfections in the electrically insulating material. These imperfections, if extant, could result in an undesirable conductive path between doped semiconductor material of a device the trench passes through and a conductor within it.
Advantageously, if this is a potential problem or concern, the alternative variant shown in
As with
Based upon the above, further alternative variants can be created having dual isolated (i.e. coaxial or coax) conductors. This is advantageous because dual conductors allow for greater contact density and can reduce cross-talk. In addition, with the dual conductor variants, as will be seen, the outer conductors are separated electrically from the inner conductor allowing them to operate at different voltages; for one conductor to operate as a electromagnetic interference (EMI) shield to protect against signal noise, or to allow the signals to propagate differentially through the structure so that lower noise data transfer can occur. Moreover, as with the single conductor approach, only one lithography defined precision etch is performed, the annular trench. As will be seen below, removal of the central material is controlled by the boundary metal and thus is not subject to process variations inherent in photolithographically defined steps or etching. Thus, even this approach is more reproducible and process robust.
Two example coax variants are illustrated in
Initially, the basic dual-conductor creation process follows the approach described in connection with
Following metalization, at least the trench 302 is filled with the electrically insulating material 500. The result of this step is shown in
Again, as shown in
Alternatively, as shown in
Otherwise, and thereafter, the approach is essentially the same as described previously. The via trench 2600, 2702 is made to a depth that extends to within the substrate 112 (but preferably not fully through it), for example, by a further etching process or through another suitable process, for example, laser drilling or ablation.
The via trench 2600, 2702 is then filled with a conductor 2802 and the substrate is thinned as described above. In the case of the first example dual-conductor variant (
Thus, in dual-conductor variants such as shown in
Based upon the above, it should be recognized that a further alternative coax variant, similar to that of
Of course, it will be understood that this further alternative variant may be unsuitable for some applications, due to the dielectric constant of silicon dioxide, silicon oxy-nitride or silicon nitride, or impossible to implement for others due to other factors not pertinent to an understanding of the subject matter described herein. Otherwise, the approach is the same as described in connection with any of the variants described above in connection with
For completeness, examples illustrating adding the optional additional thermally created dielectric or insulator 3002 aspect to the approaches of
Alternatively, the partial removal can be an inverse partial removal, i.e. the inner island is removed from the via trench inward, leaving a smaller island within the via trench. With this variant, the smaller island can serve as a post upon which a contact can be built up and connected to the metalization or conductor. Similarly, the partial removal can be a partial removal from the depth perspective, leaving a well or recess that can be used as the female part of a male/female connector or, if made conductive, can serve as an electrical contact.
Advantageously, it should now be apparent from the above that, as shown in
As briefly noted above, it is not necessary that the remaining void existing after removal of the central island of material be filled with anything at all. Moreover, in some implementations described herein there are specific advantages to not doing so.
By not filling the void 3210, 3310 in the implementation of
As shown in
As briefly noted above, irrespective of the variant used, the annulus trench described above (as well as the perimeter of semiconductor material if that variant is used) can be any closed shape. However, as an extension of the above, it should also be understood that the via trench need not be the same shape as the annulus trench nor does the width of the annulus trench have to be uniform, although in most implementations both will be the same shape, for ease of implementation reasons as well as capacitance or resistance or both.
An extension of the above applies equally to the variants that have an annulus of semiconductor material in addition to the annulus of insulator, i.e. the shape of each peripheral surface can be the same as the others or one or more can be different from one or more of the others as desired or as needed for the particular application.
In addition to the advantages obtainable, per se, from use of the above to ultimately create connections between two chips, the above approaches provide significant advantages in the area of chip, die or wafer stacking, particularly where the chip, die or wafer is pre-processed, e.g. it is fully formed from a function standpoint in that it already has whatever functional devices in terms of the transistors, capacitors, diodes, switches, resistors, capacitors, etc. it will contain created on it.
Creating vias using the annular via process provides a way to stack wafers in a manner which allows electrical conductivity and also requires little or no post-processing of the after the wafers are fused. This is highly beneficial, both on a cost and yield basis, particularly at the wafer level where two wafers are to be hybridized together or a wafer is to be populated with multiple individual chips. When putting wafers together, one of the key realizations is that the hybridized two-wafer piece (i.e. after putting two wafers together) has a much higher value than a single wafer piece (i.e. the single wafer immediately prior to hybridization). Likewise if three wafer pieces are stacked together, the value is even higher. Any post-processing that has to be done to a series of stacked dies after they are integrated adds a lot of risk because damage will result in scrapping a very high-value added piece.
Thus, the above processes provide a much better approach because all of the via processing and thinning occurs before the devices are stacked. As a result, fully stack-ready pieces are created that can just be layered one on top of another for joining (i.e. hybridization) with no additional wafer processing, via formation having been done post creation of the on-chip devices and prior to hybridization. As chips are stacked with the above approaches, while the value of the combination goes up and up, the number of steps to attach another layer is typically just one, namely—attach the next die (unless thinning is necessary and was not performed prior to the hybridization). This minimizes the risk of yield loss to expensive parts due to post processing inherent in stacking prior art where chips are stacked and thereafter, electrical contacts are created.
Thus, in contrast to the prior art, creating the vias before stacking allows for:
1) reduced or no post-processing on the stacked piece (resulting in less labor and higher yield); and
2) greater alignment tolerance (each chip only needs to be aligned well relative to the one immediately below (as opposed to stacking prior art which requires all pieces to be aligned in common relative to the bottom piece)).
Alternatively, the material 3806 can be created “automatically” during the deposition of the insulating layer 3812. For example, we have put TEOS (oxide) on the wafers by eliminating the first deposition of the material 3806, etching the trenches 3808, then depositing TEOS. Because of the way this material deposits, it placed 2.5 microns of material on top of the wafer and 1.25 microns on the walls in the trenches. This provides an alternative approach to getting a thick top layer while still covering the walls of the trenches. In other words, with this alternative, putting the material 3806 on as a separate step could be eliminated or be used in conjunction with the remaining steps depending upon the topology of the wafer.
Metal 3814 is then introduced into the trenches to provide a seed layer for plating of a conductor (
In still other alternative variants, the steps of
At this point, a generic through-chip connection has been created that can facilitate stacking on a chip, die or wafer basis and thereby form one or more multi-wafer units.
Note that, in each of the three stacks of
Depending upon the particular application in which the above will be used, the contacts can be formed in a number of ways. For example, the vias can be microbumped with, for example a C-4 solder type process of the prior art, so that two points to be electrically connected are placed into contact and the solder is changed to a liquidus state and then hardened so that the two pieces will be physically and electrically joined. In other variants, a pair of contacts can be used where one contact of the pair is rigid and the other is malleable relative to the first and a process as described herein is used to join them. In yet other variants, both contacts in the pair can have malleable material on them and an appropriate process as described herein or otherwise is used to join them. Alternatively, a post and socket type approach of the prior art can be used. With this approach, the two contacts to be joined are made in complementary shapes where either the post is slightly oversized relative to the socket or the socket is slightly undersized relative to the size of the post such that bringing the two together results in an interference fit between the two.
In certain cases, it is desirable to use thicker wafers (
The approach works with both traditional via processes where a single via is performed and insulator and metals are deposited in one hole or in our previously described process with the annular via approach to create highly controlled impedance vias.
In addition, the back to front approach can be used where one side has a non-completely filled via so that the unfilled part of the via can serve as a “slot” that will receive a “post” (i.e. a pressure or interference fit connection and thereby provide for alignment and/or physical connectivity as well as electrical connectivity. This type of pressure or interference fit aspect is illustrated in
In another alternative variant, the back-to-front method of via creation described above can be used to create a connection only part-way through the chip in such a way so that capacitive coupling can be used to send data between the chips. Because capacitive coupling works when the contacts are closer, and because the density of connections is limited by crosstalk, variants of approaches described herein are ideal for creating chips using this type of communication. These approaches readily allow for minimization of crosstalk by close connections, because it is possible for the distance between the contacts to be minimized and through use of coax or triax posts so that shielding can be provided. Moreover, capacitive contacts have the advantage that no actual electrical contact between the parts is necessary. With this approach, shown in
In addition, capacitive coupling can be used with a pressure fit connection if, for example a true back to front connection is not created in that the two vias do not linkup (i.e. material is left between the via created from the front side and the back side post). In such a case, the front side via will be independently created according to one of the variants described herein, as will the back side via.
Still further, capacitive coupling can occur between one or more contacts on a chip surface (whether created through a via approach or other approach). This may be desirable if, for example, with a stacking approach, chip heights do not allow for two complementary contacts to easily physically touch although they are close together because, for example there is a chip or metalization or other topology maintains a separation between the two, or one or both are covered by an insulator, like TEOS, a photoresist or some other oxide.
From the foregoing, the versatility of our approaches should be more apparent. Advantageously, even further variants can be created that illustrate the broad and versatile range of possibilities available through use of our approaches. One such variant, shown in
First, the initial wafer is thinned to the extent necessary to ensure that the via can go completely through the substrate (
In some cases however, the pressure fit connection approaches will not be suitable because of a lack of control-ability. For those instances, an optional alternative approach we have devised called a “post and penetration” approach can be used. Ideally, the post and penetration approach can, and typically will be, used along with a “tack and fuse” process owing to the advantages each provides alone and the further advantages provided by their use in combination.
The approach involves the use of two contacts in combination: a rigid “post” contact and a relatively malleable (with respect to the post material) pad contact, in some cases, either or both having an underlying rigid supporting structure or standoff. In simple overview, one of the two contacts is a rigid material, such as nickel (Ni), copper (Cu) or palladium (Pd) or other suitably rigid alloy such as described herein. This contact serves as the “post.” The other of the two contacts is a material that is sufficiently softer than the post that when the two contacts are brought together under pressure (whether from an externally applied force or a force caused, for example, by flexation of the wafer) the post will penetrate the malleable material (the “post and penetration” part) and heated to above a pre-specified temperature (the tack phase of the tack and fuse process) the two will become “tacked” together upon cooling to below that temperature without either of them reaching a liquidus state.
Note that, as used herein, the term liquidus is intended to mean a state in which the metal or alloy being discussed is in a fully (or substantially fully) liquid form. When a metal is in a non-liquidus or semi-liquidus state, as used herein, the metal is sufficiently soft to allow for attachment as described herein, but is insufficiently liquid to allow it to run or flow like the same metal or alloy would in a pure liquid or liquidus state. Most variants of our processes operate with the metal or alloy in a non-liquidus and non-solidus state. Stated another way, on a phase diagram for the metal or alloy, our process variants operate between the solidus (fully solid) and liquidus (fully liquid) temperatures, with most operating near the equilibrium point between the two. This difference can be further understood with reference, for example, to the joining a chip to another element as illustrated in
Thereafter, a second heating to above another temperature higher than the “tack” temperature (the fuse phase of the tack and fuse process) will cause materials from each to inter-diffuse (in contrast with a solder which would enter and exit a liquidus state (i.e. melt and re-harden)).
The tack and fuse integration process is separated into two main components: an “attach” or “tack” phase and a “fuse” phase. The tack phase makes a fairly homogeneous electrical connection between the pairs of contacts. The combination of forming a post and penetration connection with the tack process enables any surface oxide on any of the contacts to be more easily broken through. This non-oxide inhibited contact approach allows for a simpler fuse process without the need for application of significant pressure. In the absence of the combination of post and penetration and tack phase, the fuse process would require substantially greater pressure in order to allow the contacts to break through the oxides that would form at the surface of the rigid and malleable materials during the high-temperature portion of the tack process, or in the early stages of the fuse process. By getting beyond that oxide ‘crust’ at the initiation of the tack phase, the fuse phase can occur at substantially lower pressure, in some cases at no added pressure beyond the weight of the chip itself.
At this point, a further terminology convention is introduced. It should be understood that, as set forth herein, the terms “daughter” and “mother” are used, for simplicity, to generally connote whether the particular contact on a wafer being discussed is a rigid or malleable contact, with the term “mother” being associated with a rigid contact and the term “daughter” being associated with a malleable contact. Although shown fairly consistently one way herein, it is important to note that the terms “mother” and “daughter” are arbitrarily applied. Individual contacts on each wafer can be either a rigid or malleable contact as long as the corresponding contact on the other wafer to which it will be joined is of the opposite type. Thus, a given wafer surface can exclusively have one or the other type of contact or, in some variants, a single wafer side can have a mixing of both types. However, mixing of types on a single surface can be problematic for some applications and, in those applications where it is used, mixing of types on a single surface can complicate the processing unless the different types are not intermixed in one area but rather are confined to discrete areas such that large areas will contain only one type of contact allowing areas that will contain the other type to be easily protected when certain processing steps are carried out.
During the attach or tack phase of the process, the “mother” wafer is populated with “daughter” chips. The mother wafer is maintained at a single temperature (i.e. the mother wafer is maintained as an isothermal substrate during this attach process). The isothermal temperature for the mother wafer can be as low as room temperature, although raising the temperature above room temperature speeds up this phase of the process. However, the isothermal temperature is kept below the melting point of the malleable material on the daughter chip as well as the tack or the fuse temperatures. Thus, the tack process can be done by heating each small daughter chip to a higher temperature than the mother wafer so that, when the two chips are brought into contact and a post and penetration connection occurs, the interface for just that chip reaches or slightly exceeds the appropriate “tack” temperature. In general, for the primary materials discussed herein, the tack temperature would be between about 190° C. and about 320° C., with a typical nominal tack temperature of about 270° C. In this manner, the other chips on the mother wafer are not heated beyond the point where their contacts see the elevated temperature, a condition which could change the performance of the contact and cause some contacts to see much longer times at elevated temperatures than others, potentially causing non-uniformity of performance.
The tack or attach process can be performed by, for example, keeping the mother wafer at an isothermal temperature below the malleable temperature, bringing the daughter chip to the mother chip, heated to below the malleable temperature, making contact between the two chips, and quickly ramping the daughter chip temperature to the appropriate tack temperature. Thus, once the daughter chip is attached to the mother wafer, the machinery that aligns the parts (and imparts heat to the daughter chip) releases the daughter chip after applying only enough pressure to allow some contact between the parts, for example less than 2 g/contact pair, and preferably less than 1 g/contact pair.
After release, the cap/adhesion layer (or malleable layer if the malleable material also performs the function of the cap/adhesion layer) on the daughter chip becomes less soft under the decreased temperature which would be dominated by the mother chip at that point. For example, with the baseline materials described herein, the mother chip/wafer substrate can be held at between about 230° C. to 250° C., the daughter chip can be brought to the mother chip at a nominal temperature of about 270° C. and quickly ramped, after contact, to about 310° C. to 330° C. The order of the contacting relative to the quick ramp (i.e. whether it happens before or after contact with the mother wafer) can be changed. Notably, we have found that by bringing the chips into contact first and then ramping up the temperature, oxide formation on the surface of the malleable material can be minimized, thus allowing a more reproducible contact. Advantageously, through use of the malleable material, the amount of pressure per contact pair can be low. We have used applied pressures ranging from about 0.001 g to about 10 g per contact pair although lower bounds are possible, the lowest being the effect of gravity on the mass of the chip itself (i.e. its weight).
In addition, as noted above, for the tack process, daughter wafer temperatures as low as room temperature can be used if sufficient pressure is applied to break through any surface oxides. In this manner, the entire mother wafer can be populated with daughter chips before any tack phase is started. Even using this approach, due to the speed in which the process can occur, the mother wafer does not have time to be heated to any substantial degree. Thus the attachment of a second daughter chip to a mother wafer, even within 100 microns of the first chip in the horizontal or vertical direction does not soften the cap/adhesion layer of the first chip to affect its alignment to any meaningful or substantial degree.
Advantageously, the tack and fuse processes are both typically non-liquidus process. This means that the process is done so that the malleable material becomes softened significantly but does not become completely liquidus during either the tack or fuse processes. This is because if the malleable material were to become liquidus, there would be significant risk that the resultant liquid would run and short out adjacent contacts. By keeping the materials non-liquidus, far greater contact density can be achieved. However, in some variants a semi-liquidus state is allowable (i.e. some, but significantly less than all, of the malleable material may briefly become liquidus). However, those variants generally have the common characteristic of using some other type of containment mechanism to prevent the liquidus malleable material from having an adverse effect by constraining it to a defined area to avoid the possibility of shorting an adjacent contact, for example, by ensuring that the pad onto which the malleable material is applied is surrounded or covered on its periphery by a non-metallic substance into which the malleable material can not easily interdiffuse.
In some variants, in conjunction with the “tack” phase of the tack and fuse process, it may be desirable to cap the malleable material (for example, Au/Sn alloy) with an adhesion layer (for example, Sn) which will melt at lower temperature to help speed-up the tack time to enhance throughput. In addition, in some variants, it may be desirable to keep the mother wafer at an isothermal temperature of the highest possible temperature below the fuse temperature such that no degradation of a bond occurs if the chip sits at that temperature for extended times under non-controlled environmental conditions (i.e. the time it could take to populate a whole wafer in volume). We typically use 230° C. although the temperature could be higher to speed up the process. The impact of the lower temperature is an alteration of the temperature and pressure profile of the penetration phase of the attachment. Moreover, in order to speed up the process, it is desirable to have the serial processes of the tack phase (i.e. place and heat) occur as quickly as possible. A further aspect to note is that, in some variants, the longer the time spent in the tack phase, the less critical the fuse phase is for yield, etc. For example, at one extreme, on an FC 150 (for silicon-to-silicon), we had a tack phase lasting for about 1 minute and there was no fuse phase needed. This is summarized in
At the other extreme, in high volume cases, alignment would typically take about 1 second, the tack phase would take 2 to 4 seconds before the fuse phase. Thus, in those variants the environment for transport from tack machine to fuse phase can be important for getting good contacts.
Between these two extremes is a continuum of process options where the tradeoff is among 1) throughput, 2) complexity and 3) criticality of the fuse process. For very fast tack processes, the 2 to 4 second variety, the chips may be held lightly, thereby possibly requiring a reducing environment during the fuse phase or even require applied pressure of a more substantial amount during fuse. At the other end of the spectrum, the 1 minute tack process done at higher pressure and temperature, the tack itself can do a relatively good job of preliminary “fusing” of the chips. In this case the subsequent “fuse” process may merely be a contact anneal coupled with a method to ensure consistency across the wafer, and it may not require any specific environment (or pressure, if the planarity of the chip placement during ‘tack’ is adequate). This continuum is illustrated in
An important advantage to the tack phase is that, because electrical connections are non-final and easily undoable, testing of the chips can be performed after the tack process is complete but before the fuse process begins. This allows for testing and identification of bad dies both before and after this first phase of hybridization (i.e. to determine whether an individual chip that was performing before hybridization to another chip has been adversely affected by the hybridization process or is ineffective in combination with the chip to which it was attached. Moreover, the testing can be done, in the case of diced daughter chips being populated onto an undiced mother wafer, before the mother wafer is sawn or diced.
A further important advantage to use of the tack phase is that, because the chips are not combined very strongly, it is possible to easily take apart the joined chips if subsequent testing resulted in a determination that one of the joined chips was nonperforming. Separation of two chips from each other can be performed by using heat or pressure or both in combination. In the case of individually diced daughter chips being populated onto an unsawn or undiced mother waver, if a daughter chip was the problem, another “known-good” daughter chip could then be attached to the mother wafer. If the particular mother wafer chip was bad, it could be noted as such so that no further daughters would be attached and it could easily be identified immediately following dicing of the wafer, in both cases significantly increasing overall yield. In addition, if the mother chip was the one that was non-functional, the removed daughter chip could be saved for a future mother chip attach, again increasing yield and potentially decreasing cost. For example, say the malleable contact of the daughter wafers were a gold-tin or gold-silver-tin alloy and the malleable cap was tin. The tin could be attached at low temperature and, if thin enough, would not spread like thick solder balls. If a daughter chips tested bad, the individual chip on the mother wafer could be heated and pulled apart and another daughter chip attached. Once all of the daughter chips were attached and the combinations tested good, the whole mother wafer could be fused together.
Thus, the tack and fuse approach permits one only integrate known-good dies. In addition, this approach significantly reduces the risk associated with stacking of multiple dies because, a single bad chip does not require scrapping the entire stack. For chips or stacked units that are expensive, this is a significantly valuable advantage in and of itself.
In addition, the tack and fuse phases provide the additional advantages of being low pressure processes. The force used for the both the tack and fuse phases are typically less than 2 g/contact pair for contacts on a 50 micron pitch or less. At the fuse phase, we have proven use of forces of 0.8 g/contact pair to 0.001 g/contact pair. For a 400 contact chip we used 300 grams and for a 10,000 contact chip we also used 300 grams giving a range of 0.75 g to 0.03 g per contact pair; With larger numbers of contacts, e.g. 900,000, we have used use 3 Kg giving 0.003 g/contact pair. Ideally, for speed purposes, the approach uses the least possible force and, under proper circumstances, no force at all beyond the force imparted by gravity on the chip (i.e. the weight of the chip) itself.
Conventional processes for attaching dies together require attachment strengths of several grams to tens of grams per contact pair. This causes enormous stress on each of the semiconductor chips, often leading to damage or cracking. Thus, the described approach dramatically reduces or avoids imparting the levels of stress found with conventional approaches.
Moreover, more conventional approaches are not compatible with the small size dimensions that we can employ. Typical solder processes are liquidus processes and are not compatible with such small sizes and pitches nor are the pressures of several grams per contact pair. In other words, at the typical 5 g/contact pair, a chip with 10,000 contacts of 1 cm×1 cm would require 50 Kg to attach. In contrast, the pressure during the fuse portion of the process is typically less than or equal to the pressure used in the attach process. For example, using the fuse process described herein, the 10,000 contact chip that required 300 grams of pressure during the tack phase only required 9 grams during the fuse phase of the process.
In addition, the use of little to no pressure makes multiple reflows/multi-high stacks practical: In order to create stacks multiple chips high the amount of pressure on the chips should be low to prevent chipping, yield loss, the possibility of disconnecting lower chips in the stack, etc. during the fusion of chips above it, particularly, if some chips on the mother wafer might receive taller daughter chip stacks than others. If substantial pressure were needed to be put on the mother wafer and the daughter chips during the fuse process, and some of the mother chips had far larger stacks than others, a complex set of tooling could be required in order to maintain the correct pressure on each chip. In contrast, with our approach which requires only light or no external pressure, this can be avoided, making multiple high chips far more practical and allowing for stack differentials of double height or more.
A further advantage to variants of the approaches described herein is high strength following completion of the fuse process. The strength of the contacts after the fuse process is typically over several hundred kilograms per square centimeter with 1000 kg/cm2 being typical. Of course, as a result, once the fuse process has been completed, the rework potential is dramatically reduced.
Representative non-limiting example malleable materials include Gold-Tin (Au/Sn) and Silver-Tin (Ag/Sn) as well as others also identified herein. At this point it should be noted that the term “post” is one of convenience used simply to denote rigidity. It is not intended to in any way limit or mandate size, shape or geometry. Thus, as described below and in the “Specific Variants” section, the “post” could be wider than it is tall or have any cross sectional profile sufficient to accomplish the intended purpose described herein. Moreover, the “post” can be created as part of the processes described herein, for example, by thinning the back of a wafer without thinning the metallization or metal contact, or it can be created separately and attached to, or inserted into, a wafer thereafter.
Where stacking is involved, a given electrical connection through a wafer can have a rigid contact on one end and a malleable contact on the other. In such cases, for simplicity herein, once a wafer has been designated “mother” or “daughter” that term will be retained even if, for a subsequent stacking layer, the “daughter” wafer should properly be designated “mother” because the contact at issue is now a rigid contact for purposes of forming a post and penetration connection. For further clarity, a subsequent “daughter” wafer connecting to that other end will be referred to as a “daughter wafer 2.”
An example of this approach is illustrated in
As shown in
In addition, it is worthy of note that the “width” of the malleable contact can be “minimal” in that it is about the same width as or narrower than the contact (prior to joining) to which it will be connected or it could be an “extended” contact in that its width extends well beyond the minimal width. In examples above,
In general, there are advantages to making the size of the malleable contact slightly larger than the rigid contact, i.e. using an extended contact. By doing so, the malleable contact will envelope the rigid contact and the alignment accuracy between the two chips during integration can be less because, in such cases, the rigid need only penetrate somewhere within the area of the malleable. As a result, a greater alignment offset can be accommodated. This is best understood by way of example through consideration of a malleable contact of circular cross section having a 12 micron diameter and a round rigid contact of a diameter from between 10 microns and 6 microns. With a rigid contact having a 10 micron diameter, an offset of 3 microns could cause the edge of the rigid material to extend beyond the limit of the malleable. For a rigid contact having a 6 micron diameter, a 3 micron offset would still fit within the 12 micron diameter of the malleable contact material. Typically, the overall rigid contact will be less than 40 microns across at its widest point, and can be less than 25 microns, 15 microns or even less than 10 microns across at its widest point. In addition, with this approach, the malleable should be at least as wide as the rigid and preferably 20% or more wider. In addition, the post height can be greater than or less than its width, but will typically be wider than it is high.
Bearing the above basic description in mind, the approach can be extended to the variants described above by, for example, employing a suitably rigid material as one of the metalizing or conductive material so that it can be used as the rigid contact and, by applying a second more malleable material to another portion of the metalizing or conductive material so that it can serve as the malleable contact for purposes of attaching to other components or stacking.
As shown in
Thus, it will be appreciated that by using the post and penetration approach of
In addition, variants of certain of the implementations described above can be created to facilitate use of the post and penetration contact approach. For example, in implementations similar to the one shown in
First, we start with a daughter wafer
Next, a thick dielectric layer 5402, 5404 is deposited on the chips (
Then, the dielectric is etched through to provide access to the IC contact pads (
Alternatively, the thick dielectric layer 5402, 5404 could be a thick photoresist layer (
Next, a seed layer is deposited on the wafer to facilitate a later plating process (
Then, a dielectric layer is applied (
Thereafter, the wafer is plated until the desired amount of metal is present (
The dielectric is then removed, leaving “standoffs” or elevated contacts (
As an aside, in general, both the mother wafer and daughter wafer can have standoffs. On the daughter wafer, the purpose of the rigid structure is to provide a standoff to allow the overall contact to accommodate non-planarity of the two chips so that the contact can be made reliably and it may not be needed in some cases. On Mother Wafer, the purpose of the rigid structure is both as a standoff and as the post which can penetrate into the malleable material on the daughter wafer. In addition, standoffs can also be used to allow for height differences between the top IC cover glass and the IC pads so some contacts can rest on top of the glass and others on top of the pads.
Returning to the process flow, a further etch is performed in order to remove unwanted seed layer (
Next, a barrier layer is applied to the contacts on the daughter (
Again, a dielectric 6502 is applied to the daughter wafer (
Then, the malleable contacts 6702, 6704 are built up on the standoff (
The daughter wafer is then flipped and aligned with the mother wafer photolithographic patterning, the area above the contacts through which access will occur is opened up (
The two chips are then brought together under pressure so that the rigid contact penetrates the malleable contact (
Finally, the two chips undergo the fuse phase, leaving the two chips permanently attached to each other (
In addition, as shown, the spacing of the pads on the daughter wafer (
This variant proceeds as follows. First a thick dielectric is applied to the wafers (
The exposed areas above the contact and reroute route on the daughter wafer is metalized with a barrier layer (
The photolith is then stripped from the daughter wafer (
New photolithographic patterning is done to define the area where the contacts will be built up (
The malleable contacts are the created on the daughter wafer by depositing the appropriate materials, in this case, a gold-tin (Au/Sn) alloy topped by a discrete layer of tin (Sn) in turn topped by a layer of gold (Au) (
The photolith is then stripped off of both the daughter and mother wafers (
Then, the unwanted remaining exposed seed layer is removed from the mother wafer (
Finally, a cap (optionally preceded by a barrier) is applied to the mother wafer contacts to prevent oxidation (an oxide cap) (
As with previously described variants, the wafers are then aligned (
Having described several variants in more cursory overview, an additional variant will now be presented that includes further details of various steps in the process. It should be understood, however, that those details are equally applicable to the preceding variant as well as the other variants described herein.
Moreover, although illustrated in parallel form for some variants, the processing described herein need not be done in parallel and could represent different variants occurring on the same wafer or at different times on different wafers.
This example begins with the wafers 8800, 8802 respectively shown in
First, a thick dielectric layer 8902, 8904 is applied to the wafers 8800, 8802, in this case silicon wafers having aluminum IC pad contacts 8804, 8806 (
Next, a photolith layer is applied and patterned to protect the wafer from etching in undesired places (
Then, an etch is performed on the wafer (
At this point it is worth emphasizing certain attributes and advantages resulting from use of the illustrated process in certain implementations. Attributes and advantages arising from the approach include the fact that etching and creation of the vias can occur before hybridization (chip-to-chip, chip-to-wafer, or wafer-to-wafer). In other words, it is easily performed before the chip, die or wafer is joined to another element. Moreover, this approach allows for etching the vias from the device (i.e. active) side of a previously made and usable electronic chips. The approach can be used virtually anywhere on the chip where there is no circuitry directly in the path of the etch that can not be sacrificed. Thus, vias formed using the approach can be aligned with pads, or not, as desired. Still further, by making the vias over the pad and/or, in some cases, making the vias much smaller than the pad, particularly in areas of the chip where there is little or no circuitry, loss of “real estate” on the IC for circuitry can be minimized.
With respect to via formation, in some cases it may be desirable to have sloping vias in order to ensure subsequent material deposition adequately coats the sidewalls. In such cases, the slope can be a typical nominal slope of about 88 degrees off of a perpendicular to the vertical axis of the via (i.e. the via width will narrow slightly with increasing depth). A cross sectional photograph of one sloping vias example is shown in
Typically, via depths of 75 microns or greater are used having widths 5 microns or more. The vias of
Optionally, the bottom of a via can be formed so as to have a point. This is a way we have used to ensure a strong rigid post, good penetration of the rigid material into the malleable material, and a strong final contact (to maximize surface contact between the rigid and malleable materials). In order to do this we have been used an approach where the rigid post is made in a pyramid-type shape (or a pyramid on top of a cylinder) where the base of the post is as wide as the underlying contact (maximizing the strength of the attachment of the post to the contact) while the top is tapered to be much smaller than the contact, allowing the alignment relative size factor to be achieved. This variant has the advantage that it will result in formation of a pointed post and thus, when used for a post and penetration connection, will allow penetration similar to that of a later-formed pyramid-type profile of a rigid post.
Next, the photoresist is stripped (
A metal barrier layer is then deposited on the dielectric, (
Next, a plating “seed” layer is applied if metal is to be plated in the particular variant (
Both the barrier and seed layer are typically deposited by sputtering or physical vapor deposition (“PVD”), but electroless plating can be used since, for some implementations, electroless plating will provide significant advantages over sputtering or PVD. The via is then (typically completely) filled with a metal or other conductor to form the electrical conduit through the wafer (
As shown in
Advantageously, if the via is filled with the same material as the rigid material for the mother wafer or the same material as the malleable material for the daughter wafer, stacking advantages can be achieved. Alternatively the via could be filled with the same material as the malleable material if the mating contact on the chip to which it will be attached has a rigid material on it.
Note that, as shown in
Where the particular wafer is to be joined to another wafer, as is expected for most implementations, it is important that the construction of the barrier and the via filling material of a daughter wafer follow the same guidelines as the barrier and rigid materials for a mother wafer so that when the daughter chip is hybridized to the mother wafer, it performs in the same way that the mother wafer would.
Returning to the process flow, as a result of the plating in the previous step, a large amount of conductor is deposited on top of the wafer and needs to be removed. This can be achieved through lapping, polishing or chemical-mechanical processing (“CMP”). This thinning occurs down into the thick dielectric that was deposited in the first step. The actual thickness used for the dielectric applied as the first step is chosen so as to give a margin of error to this lapping step. This step can be unnecessary if the conductor filling the via is not deposited by electroplating. As shown, a chemical mechanical process (“CMP”) is then used to remove the excess plating material and underlying seed layer down to, and slightly into, the surface dielectric layer (
Next, a photolithographic etching process is again used to assist in providing access to the wafer's IC pad contacts 8804, 8806 from the top of the wafer by application of a photoresist (
Thereafter the photoresist is stripped away and the wafer is cleaned, leaving a completely formed post within the daughter wafer (
At this point, it is presumed that the wafer will be further prepared for hybridization to another element, such as another chip, a die, or a wafer (i.e. the approach is equally to all permutations of hybridization: chip-to chip, chip-to die, chip-to-wafer, die-to-die, die-to-chip, die-to-wafer, and wafer-to-wafer). This further processing is illustrated in simplified, parallel form in
The process proceeds as follows. First, a dielectric layer is applied to the mother wafer except over the IC contact pads. (
Next, a barrier layer is deposited on the daughter wafer (
As shown, a barrier material, for example, Ni/Au, Ti/Pd/Au or Ti/Pt/Au to name a few, is deposited on the daughter wafer via sputtering. In addition, this barrier can generally be used as an under bump metal (“UBM”) and for rerouting that does not require seed removal. This layer is typically put down using either a sputtering and/or evaporative process or an electroless plating optionally combined with an electroplating process for the upper layers.
In addition, as shown, a seed layer is deposited on the mother wafer (
Optionally, and alternatively, the barrier and seed layers could have the same constituents. In such cases, the single material can function as both layers.
As shown in
If, the subsequent materials can be put down with a process other than electroplating, e.g. by sputtering or evaporation, then the mother wafer steps could alternatively include patterning with lithography around the pads, putting down the barrier metals, putting down the subsequent metals and then doing a lift-off process. The net result of metals and barriers being primarily around the pads or where a reroute is desired would be the same.
Then a lithographic process is performed on the daughter wafer to expose the barrier material that is over the original contact (
This step, (
Next, the daughter wafer is metalized by depositing appropriate metals on top of the exposed barrier (
In addition, on the mother wafer, the void created by the lithographic process is filled by plating (electro- or electroless) the seed layer exposed by the lithographic process (
Upon completion of the metalization and/or plating, the photolith is then stripped away, exposing the built up contacts on the daughter and mother wafers (
Next, a photolithographic process is employed to protect the built up contacts or posts but allow for removal of the unwanted barrier and seed materials from, respectively, the daughter and mother wafers (
However, since the seed and barrier materials in this example were electroplated, an etch will be used. Thus, the unwanted seed and barrier material is etched away (
Then the photolith is stripped away (
At this point, the daughter wafer contains a functional rigid post suitable for use in forming a post and penetration fit connection with another wafer.
However, as will be evident from the description herein, in this case processing of the mother wafer continues, specifically, through electroless plating of a malleable material (relative to the material on the daughter wafer post) onto the contact (
At this point, the mother wafer now has a functional malleable post for use in forming a post and penetration fit connection with another wafer.
However, in this example, it was pre-planned that a third chip was to be stacked on top of the daughter wafer, hence the formation of the post into the wafer. Thus, further processing of the daughter wafer is required and proceeds as follows.
First, the front side (i.e. device and contact bearing side) of the daughter wafer is protected by application of an appropriate removable, protective material to protect it from contamination during subsequent thinning (
Next, the back side of the daughter wafer is thinned to expose the via fill material (e.g. the previously formed post) from the back side, typically until the daughter wafer is about 75 microns thick because the typical vias go to about 75 microns deep. If the vias extend deeper, less thinning may be required. Depending upon the particular application, the thinning is specifically done until the post extends above the back side wafer surface or, in some applications, the post will be flush with the back side surface (
In this case, because another post and penetration connection is desired, an etch is performed on the back side so that the post extends above the surface (
Of course, for daughter wafers that have no through-connections, the thinning and etching steps will generally be unnecessary subject to other height considerations making it desirable nonetheless.
With variants that use a very thick layer or carrier on the front side, thinning can potentially far exceed the typical 75 micron finished thickness. Indeed, with those variants, thinning result in a thickness down to about 10 microns. Moreover, if the carrier wafer will not be removed after a tack and fuse process, the wafer can be thinned to about 5 microns.
Note: In alternative implementations, the thinning steps can be done after hybridization between the mother and daughter. In such variants, the sequence of events would be electroless plating of the mother contact, tack, fuse, thin the daughter, etch the back side of the daughter to extend the contact above the back side surface, application of the barrier and cap to the back side contact, with the front side protection and removal of that protection being omitted as unnecessary.
A barrier and cap or cover layer is then deposited on the post (
In addition, this barrier does not have to be deposited by electroless plating. Instead, in some variants, electroplating can be used, if a seed layer is deposited on the back, plated and then etched in a similar fashion as described above. In other variants, a patterning and evaporative or sputter or other type of deposition process could be used to apply these barrier layers. While requiring more steps on a thin wafer, these alternate approaches have the advantage of being able to also define a reroute layer, shield or ground plane on the back of the wafer either through the etching of the seed layer in an electroplated process flow or through the liftoff process in a deposited metal process flow. Then the protective layer is removed from the front side of the daughter wafer (
Alternatively, if the material that is put on as either the protect layer or the adhesive to attach the carrier wafer to the daughter wafer can withstand the temperatures of the tack and fuse process, then this step can be postponed until after the fuse process is compete. This allows for greater thinning of the daughter wafer while making it possible to still handle the individual die during the tack process without cracking or damaging the chips. In this scenario, the daughter chip will typically have its circuitry face-up (i.e. away from the mother chip) with the malleable material being on the mother chip. Of course, bearing in mind that the mother/daughter convention is merely arbitrary, the opposite could be true or, in the case of certain well attachment or other variants, the malleable material could be in the via itself or even added later.
In another alternative variant, this step could be omitted entirely and the protective layer left on permanently, for example, if the vias were not formed to stack a third chip on top, but in order to allow a chip to be hybridized with circuitry facing up rather than down, for example, if optical devices are on the daughter wafer and the top carrier wafer could have built-in microlenses or other passive elements, or if the daughter and mother wafers were RF devices and it was desirable for the two electronic circuits to not be immediately adjacent to one another. Again, this would typically require the mother chip to have the malleable material on it.
At this point, presuming that the contacts described above on the mother and daughter wafers are to be mated together, it is now possible to join the respective chips. The joining process proceeds as follows.
First, the daughter wafer is flipped over and the respective contacts to be joined on the mother and daughter wafers are aligned with respect to each other (
Then, the contacts are brought together under pressure to form a post and penetration connection (
One of the key advantages to this approach to stacking is that the rigid material penetrates into the malleable material. This permits a strong bond to occur between the two wafers since the surface area between the two contacts is larger than the size of the individual contact itself. Moreover, the bond is stronger because the type of failure necessary for the two parts to pull apart would require both a delamination of the horizontal surface of the post and a shear failure of the vertical side of the post. Notably, the latter is a much less likely form of failure, so the risk of overall failure is even more remote than either alone.
In practice, the amount of protrusion is also important. Typically, at least a half-micron of protrusion is desirable. Although less protrusion can work for some implementations, the strength goes down considerably at lower levels of protrusion. In practice we have determined that, for a malleable material which has total height of 8 microns, the rigid material will typically extend 2-3 microns into the malleable; for a malleable material of 10 microns, the rigid will typically extend 5 microns into the malleable material. A general “rule of thumb” is to have a penetration of 10% or more of the thickness of the malleable contact but have it penetrate less than 90% of the way through the malleable contact.
Another key advantage is that the penetration of the posts allows for significant non-planarity of the daughter and mother chips relative to the pitch of the contacts. For example, for contacts that are 12 microns wide on a 20 micron pitch, the height of the malleable material can be fairly high, for example, up to the point where the height matches the pitch. Similarly, the planarity deviation from contact to contact can be as wide as the thickness of the malleable contacts. For example, if the post had a height of 5 microns and the malleable material had a height of 8 microns the difference in planarity from contact to contact could be as much as 8 microns. In that case, some of the posts would penetrate all the way through the malleable material and some would have less penetration.
Returning to the process flow, following penetration of the rigid contact into the malleable contact or concurrent with it, the tack phase of a tack and fuse process can be performed. As shown in
Optionally, prior to the tack phase, an underfill can be inserted between the two chips to fill the void between the two if, for example, potential rework is not part of the process and the underfill material will not be adversely affected by the temperatures used in the tack or fuse process.
At this point, the mother and daughter wafers are joined and can be tested (and in some cases if one is faulty, replaced).
Once it is determined that a permanent connection between the two is desired, the fuse phase of the tack and fuse process is performed (
Optionally, if not done previously, an underfill can be inserted between the chips prior to the fuse process, if temperature is not a concern, or following the fuse process. The advantage to using an underfill is that it reduces the prospect of air being trapped between the two chips and later damaging the chips or connections due to temperature cycling (because the tack and fuse process forms a hermetic seal).
Once the mother wafer has been populated in the tack process (i.e. in a die-to-wafer process, the alignment and tack processes are repeated for each good location across the mother wafer, with known bad mother die sites not being populated, and in a wafer-to-wafer process, the two wafers are tacked together in their entirety, and if optional testing is performed, the location of bad chips being noted for future elimination), then the entire mother wafer goes through the fuse process, permanently attaching all of the daughter chips. This can be done at a much higher temperature than the tack phase. Moreover, the time is the same for each chip, as the process is done wafer-at-a-time, so the process yields fairly homogeneous connections across the each individual chip.
Temperature for the fuse phase will typically be, for example, 320° C. to 400° C., depending upon the particular materials involved.
Advantageously, by separating the tack process from the fuse process, the equipment performing the tack is not slowed down by having to heat or cool each individual part). By performing this in a controlled manner at a wafer level, all contacts can have a very similar final composition.
An inert or reducing environment can be used during the tack phase, the fuse phase, or both, to help to minimize or remove oxide at the surface of the materials and help lower the required temperature or pressure for each step. Typically these would be gases such as nitrogen, argon, other inert gases or reducing gases such as forming-gas or formic acid, or some other environment with hydrogen content or some other reducing gas.
As noted above, the process is not complete because a third chip is to be joined to this newly formed unit. As with the joining of the mother and daughter chips, the unit can be joined to another chip. Thus, as shown in
Advantageously, because of the prior processing steps, the exposed side of the via on the top of the first daughter chip has the same composition as the top of the original rigid contact. Thus, for subsequent “daughter” wafers, the hybridization happens the same way as done for the first two wafers (i.e. align, penetrate, tack (optionally test) and fuse. The malleable material is pinned between the respective barrier layers and the post on the via penetrates into the malleable material). An important advantage to the process is thus, that the vias and the base hybridization are set up to operate with the same material system and the same process flow facilitating repeated stacking beyond the conventional stacked chip pairs one might find.
As a result, a mother wafer can be populated with one set of chips and then another (Daughter Wafer 2), and then another, etc running the process in an identical fashion with each layer as needed using either a tack, fuse, tack, fuse approach or, in some cases, a tack, tack, tack, fuse all approach.
Thus, a second tack phase is performed on the second daughter wafer to bond it to the unit and, once competed, this newly formed larger unit can optionally be further tested and, if the second daughter chip is bad, it can be detached and replaced (
Finally, when a permanent connection between the second daughter and the unit is desired, the fuse phase of the tack and fuse process is performed again (
After this step, the process can be repeated over and over to allow multiple further chips to be integrated, for example onto the “Daughter Wafer 2” or other chips present on the wafer (not shown). Because the electrical connections are made during each tack process, each chip needs to be aligned only to the one immediately below it, so a further advantage is achieved in that there is no accumulation of alignment errors as in other stacking techniques where all of the chips must first be stacked before an attempt at through-connection can begin.
Moreover, to the extent necessary, testing of each larger combined unit can be performed following each successive layer (and rework can be done, if required). Again, this provides a distinct advantage and dramatic cost savings and yield increase because, if dies were stacked in multiple layers, conventional techniques would likely require that the entire built up unit be completed before electrical testing could occur. Thus, only after an expensive unit has been created could a conventional part be tested and, if bad and rework were not possible, the only option would be to scrap the entire high cost unit. In addition, with conventional techniques, the risk of damaging the unit during build up or of wasting parts, for example, if the failure was on the first tier chip, dramatically increases.
In contrast, using one of the approaches described herein, a multi-stack configuration could be created with much less risk. Again, depending upon the particular case, the approach could be performed, as above, as a sequence of align, tack, fuse, align, tack, fuse, as many times as was necessary. Under conditions where the tack process had high enough strength, for example >=500 contacts, then the process could alternatively be performed as align, tack, align, tack, as many times as necessary and only after all of the chips were stacked vertically (and tested good if that option was used) would the fuse be performed. This second approach can further be effectively used when different numbers of chips will be stacked at different locations.
At this point it is useful to note that, through use of the post and penetration connections and the tack and fuse processes, the subsequent joining of the second daughter wafer (and subsequent wafers) to the unit can be performed without adversely affecting the prior-formed inter-unit connections. Indeed, we have surprisingly found that by using a tack, fuse, tack fuse, approach (whether or not intervening thinning occurs), the successive fuse steps actually cause the resistance of the previous connections to go down. This is significant because common wisdom would tend to suggest that a subsequent fuse would tend to weaken or degrade prior formed connections (this was especially true with the “well” connections described below).
The process begins, starting with the wafer of
Then, the wafers are aligned relative to each other (
Now, since this example also involves addition of a second daughter wafer on top of this daughter wafer, the process proceeds as follows. First, the back side of the daughter wafer of the combined unit is thinned to expose the previously-formed back side contact (
While this adds other steps post-hybridization, namely those involved with the thinning, if this is sufficient for the particular application the process can stop here. The advantage to doing so is that no further lithographic patterning or material deposition which each require more touch labor and are the major sources of yield-loss risk. Alternatively, if the time lag to joining to another element, the material, or other factors are such that oxidation could be a problem, a cap could be added (i.e. further processing would be required).
Presuming that oxidation could be a problem, a cap is applied to the upraised portion of the post (
As with the first daughter wafer, the next daughter wafer is aligned over this back side contact (
In general, there are numerous materials that are suitable for use as the barrier. Such materials include, but are not limited to: Ni, Cr, Ti/Pt, Ti/Pd/Pt, Ti/Pt/Au, Ti/Pd, Ti/Pd/Au, Ti/Pd/Pt/Au, TiW, Ta, TaN, Ti, TaW, and W.
Suitable materials for the seed layer include, but are not limited to: Ni, Cu, Al, Au, W, Pd, and Pt.
Alternative suitable materials include, but are not limited to: Ta/Cu, TaN/Cu, Ni/Au, Ni/Cu, Ti/Pd/Au, Ti/Pd/Cu, Chromium, conductive epoxy that can be put down in a flat manner (e.g. through evaporation or spraying), or combinations thereof.
Note however, that all of the barriers on a chip or chip pair do not have to be of the exact same materials.
In general, where a barrier is used the material should have the following characteristics:
i) It should be compatible with the particular pad material (typical pads are Aluminum, Copper, and Gold);
ii) It should to be selected so that, if a wafer has coexisting small (<15 um) and larger (>50 um) IC pads, it can be placed onto that wafer with good yield on both; and
iii) If an under bump metal is also used as the rigid material or acts as a standoff, then it should satisfy the above and also be able to be made several microns (>3 um) high.
In addition, it is desirable for the barrier material to be compatible with deposition on top of both IC pads and the top cover glass/passivation layer of the chip.
The use of a barrier also can provide one or more of the following advantages:
i) It can allow for high yield and increase reliability of the contacts for hybridization;
ii) If deposited both on top of the pads and the top cover-glass/passivation layer of the chip, the barrier layer can later be used as:
1) a signal reroute material,
2) an electrical shield between the two chips to prevent crosstalk between them, and/or
3) a seed layer for any subsequent steps which can be performed by electroplating (e.g. formation of the rigid post and application of the malleable materials);
iii) increased shelf-life of the daughter material because the barrier acts as a cap to prevent or retards oxidation;
iv) it can be pre-patterned to act as a reroute or a shield;
The alternative materials noted above can provide certain advantages in some implementations because:
i) the barrier capabilities of Ta & TaN are believed to be superior to that of TiW,
ii) a nickel-based process allows the UBM and subsequent rigid material to be one and the same, thereby simplifying the process,
iii) alternatives which do not leave copper exposed have longer shelf-life so they can be more compatible with certain manufacturing processes,
iv) if no subsequent electroplating steps are needed (e.g. for deposition of a rigid or standoff member on the daughter wafer), then any of these materials could be just patterned over the pads and reroute or shielding areas, thereby eliminating the need to perform a subsequent seed and etching step to define these regions.
With respect to the use of barrier layers, in many variants, it is important to ensure that: 1) the appropriate metals that are supposed to interact do interact; 2) those same metals interact in a way that the final composition after interaction is correct, 3) other metals used in the stack (i.e. rigid and standoff) do not interact so as to contaminate the metals, and 4) the barrier will allow for multiple high temperature cycles at temperatures up to and above both the package solder conditions (e.g. Pb/Sn at the appropriate temperature or some lead-free solders operating typically near about 240° C. to about 270° C.) for the tack part of the process and temperatures for the fuse part of the process which can typically be between about 300° C. to about 350° C. The barrier maintains the integrity of the attachment material by preventing intermixing of metals that should be kept separate for better integrity of the bond.
This is shown, by way of example, with reference to
In contrast, for example, if the Nickel barrier layer 14008 were absent, then the Au/Sn 14102 would be in direct contact with a very thick layer of copper 14010 (in an actual implementation of the example, it would be over 60% of the thickness of the Au/Sn. As a result, under temperature, the Sn would diffuse into the copper and then the resulting alloy would begin to change dramatically in properties. For example, copper has a melting point of 1084° C. As the Sn initially diffuses into the copper, the top of the rigid post would be a Sn-rich mixture which would have a melting point much lower (e.g. a 97% Sn 3% Cu mixture has a melting point around 230° C.). As the Sn diffuses further into the copper, it winds up having a lower melting point than the Au/Sn and the copper post ceases to be the rigid member in the tack and fuse process. Equally importantly, the copper 14010 would leach Sn from the Au/Sn 14102 resulting in an increase in the temperature at which it becomes malleable. Thus, an increasingly softer rigid member would be trying to penetrate into an increasingly harder malleable member. This would affect contact strength, uniformity, and ultimately the density of the contact spacing that could be used. Moreover, the effect would be cumulative with time. Depending upon the length of the time over which the fuse process occurs, the composition and the performance of the contact could vary greatly. This would also be the case if a contact underwent multiple fuse cycles, for example if chips were stacked multiple-high vertically. The bottom chips in the stack would have vastly different and inconsistent behavior relative to later fused chips in the stack. By using the barrier metals, the Au/Sn is largely confined and thus can maintain the same composition and characteristics through multiple fuse cycles. Note that even with a barrier some interdiffusion can occur, for example between the Au/Sn and the Ni, but the rate of this diffusion is far, far slower than would be the case with the Cu so it can be neglected for up to a reasonably large number of stacked chips—e.g. 100 or fewer. Thus, whatever materials are used for the particular implementation, the barrier should typically be a constituent of the final joining alloy to avoid or minimize adverse interdiffusion.
In the general post and penetration approach, the two mating contacts have been shown as largely flat—although this is neither a requirement nor a necessarily desirable configuration for all applications. Since the quality (or lack thereof) of an electrical connection between two points directly affects the resistance of the connection, and poor connections reduce yield, minimizing of poor connections is desirable. Advantageously, the post and penetration approach can (without increasing the “footprint” of either contact) be readily adapted to reduce the risk of a high resistance connection being created, thereby increasing yield. The approach involves improving penetration and increasing the contact surface area by creating a pattern or profile on the malleable or the penetrating contact.
Where the relative sizing makes the malleable contact larger than the rigid contact, if the malleable contact will be directly over the IC contact pad, the malleable contact can be profiled almost automatically. By patterning the metal for the malleable contact in an area that is larger than the opening for the IC contact pad on which it is built, a natural depression will be formed near the center of the contact, due to the relative height differential between the cover glass on the IC and the IC pad itself.
Profiling the rigid contact reduces the initial contact area thereby effectively increasing applied force per unit contact area which improves penetration, while the increase in surface area afforded by the walls of the profiles in the depth direction ensures that sufficient area of electrical and mechanical contact is achieved.
For purposes of illustration, representative, non-limiting, illustrative examples of some of the myriad of possible mother contact profiles are shown in top and cutaway side views along section lines A-A in
Other alternatives can use “wings” at the base of the contact, such as shown in
Still further, it may be desirable to use an asymmetrical or elongated contact (i.e. different widths in different directions in order to absorb strain in a particular direction such as shown in
In addition, the contact profiles could include undercuts such as shown in
It should be further noted that the particular shape of the contact pad, or the shape or configuration of the profiling used is, per se, irrelevant—the important aspect being the use of some profile to increase the available contact surface area while providing an appropriate shape to bond to for the particular application, not the particular contact or profile shape used, subject to the engineering requirement that the total current requirements for the contact can be handled by the minimum acceptable amount of contact and the particular profile used increases the surface area by an amount sufficient to likely achieve the desired objective relative to the likelihood that a bad connection will result if profiling is not used. Moreover, although discussed in connection with a rigid/mother contact, malleable/daughter contacts that are analogously profiled could be used. However, in that instance, the contact configuration would most typically involve a rigid well configuration on the mother wafer.
The above is briefly illustrated with reference to
Stated another way for purposes of explanation, presume that, if the rigid contact 14604 was not profiled, the contact area would have been equal to the minimum contact area possible to meet the total current requirement for the contact. In such a case, if any portion of the contact did not result in a good connection, the connection would likely be unacceptable and could result in premature failure during use or complete unusability. In contrast, in this example, the rigid contact of
Alternatively, a profiled contact can be created by using multiple small rigid contacts in conjunction with one or more larger malleable contacts to create a single, overall connection. For example, one could have an electrical connection made up of three sets of contact pairs where each individual contact pair is made up of multiple rigid contacts and a single (or multiple) malleable contacts.
A further variant of the profiling concept involves the creation of a “well” designed, depending upon the particular implementation, to assist or improve alignment, constrain the malleable material, or assist in forming a good connection. As shown and described in connection with the following figures, these well-attach variants provide further benefits and advantages to particular implementations.
As a result, the daughter wafer's well will constrain the bonding material (e.g. the covers and malleable materials) during the penetration process, as well as during the tack (
Advantageously, through this approach, the well can allow the cover or cap materials and or the malleable material itself to be of a material that can be brought to a semi liquidus or even true melting point or, at least to a point where it becomes flexible enough so that that it would ordinarily spread. This is useful for situations where contacts are positioned close together and the flexing that typically occurs during melting will cause the materials to bulge laterally in an effort to reduce surface area. For contacts where the spacing between edges of the contacts without the wells is less than or equal to about 3 times the height of the malleable material, pre integration planning for such use may be desirable (e.g. if the malleable material is 8 microns high and the separation between the contact edges is less than or equal to about 25 microns this approach should be considered).
In addition, if brought too close to their melting temperature, some materials can “wet” a wafer surface and, rather than just spreading, they can creep along a surface. In the case of a malleable contact, left unaccounted for, such action can cause electrical shorting between adjacent contacts. Advantageously, by keeping these materials in the well, any wetting creep will be counteracted by surface tension and keep the material in the well; preventing it from shorting adjacent contacts.
The well can also be critical in some implementations, for example, if a post-joining process will be performed that could cause the combined contact to melt. For example if a contact were made at the appropriate temperature for the rigid-malleable contact to be made, and then the combined chip needed to be soldered into a package but the temperature required for the soldering step was higher than the melting temperature of the contact as it exists at completion of the fuse phase, then the contact would stay in-tact during the process because the melted material would be encapsulated by the well and re-attach upon cooling.
Moreover, the well approach is well suited to making multiple densely packed connections because the wells are patterned using semiconductor lithography techniques rather than conventional mask printing or soldering techniques.
In alternative variants, a “reverse” of the well process described above can be used. In these variants, the process is performed so that the well is not filled with the malleable metal. These variants fall into one of four classes respectively illustrated in
Class I (
Class II (
Class III (
Class IV (
Advantageously, with the approaches described above the well can either be built up using, for example, a dielectric or it can be recessed (i.e. made by etching into the semiconductor). Still further, the well can be a byproduct of the via formation process. For example it can even be part of a via that is not completely filled.
As with the other class II variants, this variant violates the daughter/mother convention because the wafer 16402 bearing the counterpart to the wafer of
A further alternative well attach variant can be formed using the profiled contacts of
1) where it is undesirable to place a cover material over the malleable material because it could adversely affect the way the materials bond;
2) where the attachment would like to be done at a very low temperature (or, in some cases, even room temperature) to enhance the speed of the process, for example, if the wafers each had very flat surfaces, then van der Waals forces could attach the chips or dangling atomic bonds could create covalent bonds allowing connections to be made by insulators such as oxides, nitrides or other dielectrics (this avoids or reduces waiting time for parts to come up to temperature and potentially decreases the cost of capital equipment since a machine with temperature capability would not be necessary); and
3) where it may be desirable to have the attach materials reflow (turn liquidus) in order to self-center the chips for the subsequent fuse process without having the primary contacts turn completely liquidus since, as noted above, that could cause running or creep and thus would limit the potential density of the actual contacts (this also allows for cheaper equipment to be used to do the attachment since that equipment would not have to have the alignment accuracy necessarily needed by the high pitch of the primary contacts because the remote attach contacts could indirectly provide that level of accuracy.
By way of example, the remote contacts 16802, 16804 could be made out of a material like indium, which is soft at room-temperature and thus, could be made to attach merely by the use of pressure squeezing the parts together. Alternatively, some other low temperature material could be used which can provide adhesion without high temperature, the particular material being largely unimportant, provided it does not adversely affect the whole (i.e. introduce shorts, etc.). For example, a lower temperature solder (less than 250° C.) could be used. If put into its liquidus state, the surface tension could align the two chips together so that the attach process can be done by cheaper pieces of equipment that have poor alignment accuracy, for example, a conventional pick-and place machine. Still further, the remote contacts could be configured so that, if very flat, simple covalent bonding aligns and holds the chips together.
In this process, as shown in
In general, and as with the tack phase, the fuse phase would occur at a higher temperature and/or pressure than is required for the attach or adhesion phase of this variant.
Again, as with materials that could turn liquidus or semi-liquidus during the tack and fuse phases, compression of the attach contacts can cause them to spread laterally and/or heating of the material could cause it to turn liquidus and want to spread out, potentially causing electrical shorting if it spread to the primary contacts. Thus, one advantageous option is to apply the principles of forming “well”-based electrical contacts described herein to the remote contacts. In this manner, they can be allowed to become liquidus or extend laterally during pressure application or at temperature during the tack or fuse process, without contaminating or shorting out the primary contacts.
Advantageously, the remote contacts can also be configured to enable testing of the two chips before bonding the actual contacts together, irrespective of, or prior to, joining in the tack and fuse phases. If the chips are designed so that the location of the remote contacts are also the location of special pads that allow communication between the chips to occur, for purposes of testing whether the combination of the particular individual chips is operational, then if either or both of the chips were not operational (i.e. non-functional or functional but out of specification), the chips could be pulled apart and a new, chip attached.
Moreover, through proper design, this pre-tack, pseudo-hybridization testing approach can be very valuable since it can be incorporated into a design, whether joining will occur on a wafer-to-wafer, chip-to-wafer or chip-to-chip basis. Thus, the selection of the type of joining to be used for a specific application (i.e. wafer-to-wafer, chip-to-wafer or chip-to-chip) can, in part, be a factor of ability to test. For example, if testing is possible on a wafer basis, then all of the chips on two wafers can be hybridized in parallel on a wafer basis, with non-operational chips being flagged for rework once sawn or diced. Alternatively, the approach can be used in cases where individual dies come from one or more foundries and there is no good way to know before hybridization if any given die is a known good die.
In yet another an alternative version, the remote materials could be of the same materials as the primary contacts (e.g. rigid & malleable) as long as they were taller than the primary contacts so that, during the initial attach phase, they did not allow the primary contacts to touch. Then during the fuse process, the remote contacts would be compressed further than the primary contacts. Advantageously, by using the same materials on the remote and primary contacts processing is simplified.
From the above discussions, a derivative variants can be derived that build upon and combine concepts from the multi-axial through vias, well attach, profiled contact and remote attach variants.
The first group of variants involve complex contact shapes (i.e. contact shapes other than the conventional single square or single dot). One such example, involves creation of a shielded contact, in the simplest case, similar to a cross section of a square (FIG. 172A) or round (
In the case of a coax, triax or other regular or irregular shaped contact, one of the contact(s) would be connected so as to be signal carrying while another would act as, or connect to, a ground plane. When used with the coaxial via (
Using the outer ring of a contact as a ground allows for shielding between the chips because, the only area where a signal propagates is through very small openings in the shielding layer. The same is true for triax connections where differential signal pairs can exist within an outer ground plane. Thus, such contacts are particularly well suited for chips carrying high-speed or RF signals.
The second group of variants center around using the contact approach for making a hermetic seal between two chips (or between a chip and a package or board) to protect connection pads, for example I/O pads, or other devices (e.g. optical devices) which might exist in between the two outer devices. In this situation, the connection pads and/or optical devices are pre-existing or concurrently are brought into existence in and will be sandwiched between two elements (e.g. two chips or one chip and a package or board). A ring, which may or may not extend wholly or partially into the chip) is formed on the two elements outside the area to be protected and configured to be joined using either a malleable/rigid or well attach process so that when the two elements are hybridized together, they form a hermetic, metallic seal around everything within it. This hermetic package can then withstand most arbitrary environments, since metal's non-porous nature renders it is impervious to most environmental conditions.
A key advantage to some variants of our approach is that, because they use malleable and rigid connections (versus other connection approaches such as a metallic solder which becomes liquidus), the connections can take on any of a variety of geometrically closed shapes. This is in sharp contrast to a liquidus material, which would tend to run and reshape through surface tension into the lowest surface area available (e.g. cubes turn to spheres, corners get rounded etc.) and, while techniques could be used to cause the liquidus material to be wicked along a pre-specified surface of the chip through, for example, capillary action, there is no way to reliably ensure proper distribution of material about the contact, avoid creating voids or prevent some of the material from running out of its specified area and potentially shorting out contacts, when complex shapes are involved. To the contrary, with variants of our approaches, the simplicity or complexity of the shape is largely irrelevant because the approach is the same irrespective of shape—the only limitations being tied to the ability to photolithographically define the shape and deposit the appropriate metals.
At this point, the rigid/malleable contact variants as well as the via formation variants can be summed up in chart form using the charts of
Similarly,
Numerous examples above have described the approaches with reference to the alternatives of depositing metal on a daughter wafer or plating of a daughter wafer. To aid in understanding,
The process begins with the respective daughter and mother wafers of
In contrast, the process flow for the plating case is shown in
At this point, the dice, align, tack and fuse processes can be preformed as desired to join the two together.
Based upon the above it is useful to note the advantages and disadvantages to each approach, which will aid in the selection of the process style to use for a particular application.
The deposition approach for the daughter wafer has the advantages of: no seed layer, no electroplating, it is a one mask process, and automatically having the compositional accuracy of Au/Sn. However, the approach has the following disadvantages: thickness control from run to run can be difficult, “wings” of metal can appear if the directionality of the deposition is off, and it may require an Au reclamation program.
The plating approach for the daughter wafer has the advantages that: the cost is lower and there is no need to do reclamation, it can be supported by major equipment vendors because conventional, currently available plating equipment can be used. However it has the disadvantage that the compositional accuracy required is +1.5%/−2.5% and potentially requires an additional mask step.
With the mother wafer, there are essentially three process variants:
1) electroless plating (illustrated in
2) thin resist electroplating process with copper (illustrated in
3) thick resist electroplating process with copper (illustrated in
The attendant advantages and disadvantages of each are as follows. The advantages of the electroless approach include: no separate barrier deposition; no seed layer deposition; no seed etch needed; and maskless Process. However, electroless plating of nickel is more difficult to control in terms of thickness or nodule formation which can affect yield and therefore may not be suitable for high volume wafer yield. The advantages of the thin dielectric process include: thinner Ni is used, so the process is more controllable; copper places lower stress on the IC cover glass; use of copper is more mainstream; and electroplating of copper can be controlled better. However, the penetration of Ni/Au onto mushroom-shaped sidewalls can be inconsistent, potentially leaving some copper exposed; a mushroom shape is not optimal for the tack process and additional process steps are required (i.e. seed deposition, seed etch, etc.).
The advantages of the thick dielectric deposition process include: better contact or “bump” shape, full copper coverage by the barrier/cap, better control of uniformity and shape, lower Ni nodule formation, rendering it typically the highest yield process in volume. However, this approach potentially requires an extra mask step if a self aligned seed etch is not effective, so this approach may require a spray etcher.
In keeping with this discussion of deposition and plating variants, some further specific details of some mother and daughter contacts are now provided to provide a further understanding of the process.
It should be noted that the ranges for the Au/Sn presented in connection with
Having now described numerous through-chip connection variants and applications relating to the electrical aspect of various interchip connections, an additional alternative optional variant that takes advantage of implementations involving an unfilled inner trench or void or variants that do not expressly involve chip to chip signal transfer can now be presented.
In particular, alternative advantageous stacking variants can be created if the innermost voids are left unfilled. By sealing the voids from the surrounding parts, but leaving them open to each other, those voids can be used, for example, to aid in cooling a stack of chips.
With this variant a series of wafers having such vias are stacked in a way such that the material at the periphery of the vias protect the via sidewalls within the resultant semiconductor wafers and creates a continuous, contiguous air and liquid-tight tube when they are attached together. The stacked pieces are arranged so the tube extends through some or all of the stack. An end of the tube through the chip stack is covered by a construction which has a condensing region, for example, they further connect to a tube embedded in a heat sink. When filled with the appropriate fluid (and a wick if necessary), each of these tubes can act as a heat pipe, pulling heat away from the IC stack more effectively. Optionally, an electrically isolated metal can connect to, and extend outwards from, the heat pipe (like fins or plates) in between the stacked chips on unused chip real estate so as to further increase the heat transfer capabilities. Moreover, such fins or plates can be formed by the barrier or seed layers, potentially allowing them to play multiple roles, for example, by acting as a shield or ground plane and a fin at the same time allowing them to serve multiple roles.
This is accomplished, for example, as shown in
Depending upon the particular implementation, one end of the tube can be sealed to the doped semiconductor material or substrate 21612 within a chip (i.e. the tube does not go all the way through) or to surface material of another chip that does not contain a portion of the tube itself but merely acts as a stopper or plug. In addition, multiple tubes can be formed with each having a different working fluid or different pressures for the respective working fluids (whether the same or different) such that they have different vaporization and condensation temperatures. In this manner, a broader range of heat pipe operation can be obtained. Still further, those heat pipes can be grouped or dispersed about the chips relative to thermal “hot spots” on the chips.
In some variants, the wick 21608, if present, can be made of, for example, a porous or capillary structure, a scintered powder, a grooved tube, a mesh, a carbon nanotube structure, graphite or any other suitable wick material. In addition, the working fluid can be any heat pipe fluid, provided that it will not corrode, degrade or otherwise adversely affect the surfaces (i.e. doped semiconductor, substrate, insulator, conductor metal, etc.) with which it will come into contact. Typical working fluids can include water, an alcohol, acetone or, in some cases, mercury. In addition, in some variants, a material that is a solid at 1 Atm (101.3 kPa) and 68° F. (20° C.) can be used if it will vaporize or sublime in a suitable manner to provide the requisite transfer of heat of vaporization required for a heat-pipe. Finally, it should be noted that a pre-formed (i.e. previously fabricated) heat pipe can be used if it is of suitable dimensions for insertion into the inner via.
Advantageously, because this approach places the heat pipes closer to where the heat is generated and such heat pipes can be interspersed throughout the chip, the approach can increase the effectiveness of whatever cooling methods would additionally be employed. In addition, it should be understood that the above approach can also be used to create heat pipes within chips where no electrical connections are desired or required.
Often, there is a desire to electrically isolate chips from one another to prevent electrical crosstalk. In addition, when stacking devices vertically to take advantage of one of the via processes described herein (or a variant thereof), there may be applications where it is desirable to connect two chips together with a third chip which communicates with both of them, may intervene between communications among them or both. As should be appreciated from the preceding, the processes for forming inter-wafer connections, although illustrated involving one or two contacts, are independent of the number of total contacts and locations of where the mating chip contacts for the rest of the wafer reside (i.e. on one or more chips). This means that, in some cases, a single daughter chip can span two or more mother wafer chips or a “daughter wafer 2” chip can span two daughter chips or a mother and daughter chip. Thus, spanning is a straightforward application of the process of adding of a “daughter wafer” or “daughter wafer 2”, the process being the same, but the full set of connections to which the daughter chip will connect do not all have mates on the same chip. However, in certain cases of this variant, the two base chips (i.e. chips to be spanned by a single chip) may be of different heights. Thus, there is a need to deal with such a height differential. Advantageously, further variants of the via processes herein allow this to be achieved.
As noted herein, stacks can be formed an arbitrary multiple number of elements high. However, depending upon the particular instance, in some cases the effect and geometry of the stacking needs to be considered in addition to the decision of whether to join in a tack, fuse, tack, fuse approach or a tack, tack, tack, overall fuse approach. For example, in a wafer scale stacking process such as described herein using through via connections, a decision must be made whether to pre-thin the original daughter wafer before it is diced for joining with the mother wafer or whether it should be joined to the mother wafer (on a per chip or entire wafer basis) and then thinned. The difference is as follows. The tack, fuse, thin, tack, fuse, thin approach has an advantage in that it eliminates a few steps and, more importantly, eliminates the handling of very thin wafers if they are thinned before dicing and joining which can detract from the yield. The disadvantage is that it requires more touch labor on hybridized parts—thinning on a more expensive hybridized part versus just the daughter wafer(s) (detracting from yield).
Another disadvantage occurs when there are several daughter stacks on the mother chip when each stack has different numbers of chips. Placement and ordering of thinning becomes important because a separate thinning step needs to happen for each layer of chips on the mother wafer. As a result, without proper planning, a point will be reached where some stacks cannot have additional chips added because they will be below the height of an adjacent stack, rendering thinning of that chip difficult or impossible.
In contrast, thinning prior to joining has the advantage that it can always be performed, however its disadvantage, noted above, is the increased risk associated with having thin wafers.
Having described numerous different alternative, optional and complementary variants, an example application of the above is now presented with reference to
Still further advantages can be achieved if the chips are designed with the likelihood of stacking in mind. For example, in the example of
As can be seen from the immediately preceding discussion, further outgrowth of the processes and aspects described herein is the ability to efficiently create different kinds of “packaging” than previously used (
At present, complex integrated circuit chips are created and packaged as shown in
By using aspects described herein however, different types of packaging can be used to advantageously aid in optimizing cost, time to process, and risk of low yield, to name a few. For example, through use of aspects as described herein, configurations such as illustrated in
Another alternative approach is illustrated in
Yet another alternative approach is shown in
A further alternative approach is illustrated in
Yet a further approach, the most sophisticated of the approaches, is illustrated in
Thus, all of the approaches of
As a result, low performance circuitry can be designed on one chip and high performance chips can be designed for higher performance technology. Moreover, this type of approach can be more cost efficient because a significant amount of high-speed technology real estate can be saved by moving low-speed circuits “off-chip” without needing powerful signal driver circuits to do so. Some examples of the myriad of possibilities are shown, in connection with a high-level representation of the processes described herein, in
At this point, some further discussion of portions of the aspects described above will be detailed. At present, in order to create an electronic chip, a wafer has to undergo two sets of processes—front-end processing and back-end processing. In front-end processing, the actual devices, including transistors and resistors are created. This involves, in the case of a silicon chip, for example, growth of silicon dioxide, patterning and implantation or diffusion of dopants to obtain the desired electrical properties, growth or deposition of gate dielectrics, and growth or deposition of insulating materials to isolate neighboring devices.
In back end processing, the various devices created during the front-end process are interconnected to form the desired electrical circuits. This involves, for example, depositing layers of the metal traces that form the interconnections, as well as insulating material, and etching it into the desired patterns. Typically, the metal layers consist of aluminum or copper. The insulating material is typically silicon-dioxide, a silicate glass, or other low dielectric constant materials. The metal layers are interconnected by etching vias in the insulating material and depositing tungsten in them.
At present, for a 12″ wafer, using 90 nm processes, front- and back-end processing each take about 20 days to complete, and they occur serially. As a result, it can take more than 40 days to fabricate a single wafer from start to finish.
Advantageously, using the processes described herein, that time can be cut to nearly half for most current submicron design rule based chip fabrication technologies (for example, 0.5 μm, 0.18 μm, 0.13 μm, 90 nm, 65 nm, 45 nm, etc. . . . ) because the above approaches can allow front- and back-end processing to occur concurrently, in parallel and even in different and unrelated foundries. This is accomplished by performing the front-end processing in a conventional manner on one wafer (a front-end or “FE-wafer”) and back-end processing in the conventional manner on another wafer (a back-end or “BE-wafer”), in parallel, as if the two were the same wafer. In this way, the routing can be performed in a cheap foundry, relative to the transistor or other device-bearing portion, and each can be created in about 20 days. Then, by thinning the wafer and creating connection points on the back side of the FE-wafer through use of one variant of the via processes described herein, connection points can be established thereon. In a similar manner, the processes described herein can be used with the BE-wafer to create a set of complementary connection points corresponding to those on the FE-wafer. Thereafter, the two can be joined together using, for example, a tack and fuse approach, if malleable and rigid corresponding connections are formed (typically with the FE-wafer being the daughter wafer of the above processes (i.e. carrying the malleable contact), a remote attachment approach as described herein, covalent or other wafer surface bonding techniques (alone, with a through-via approach, and/or with simple filled vias that serve to lock the two together and maintain alignment, or some combination thereof/alternative thereto.
Advantageously, through this approach, the metal layers do not have to be limited in thickness or density as might be required by the topology and stress limitations imposed by ever increasingly sensitive transistors. In addition, by separating the process into two chips, lines can be larger and there can be more layers, thereby potentially allowing greater in-chip connectivity and lower parasitic resistance for faster cross-chip communication.
Advantageously, because our approach is independent of the particular fabrication or interconnect technology used to create the particular FE-wafer or BE-wafer, or the design rules applicable to such fabrication, the processes described herein can be used to bring dissimilar technologies together at the nano-level. In other words, the approaches described herein are independent of whatever chip design rules are appropriate to ensure that devices or their interconnections do not overlap or interact with one another in undesirable ways for the particular material (a Si wafer, a GaAs wafer, a SiGe wafer, a Ge wafer, an InP wafer, an InAs wafer, an InSb wafer, a GaN wafer, a GaP wafer, a GaSb wafer, a MgO wafer, a CdTe wafer, a CdS wafer, etc.), or what high resolution mask or non-mask based approaches are used to form submicron or sub-nanometer features or define spacing between devices, their interconnections, or the geometries of the interconnections themselves. Thus, the advance described herein allows for chip fabrication technology to shift from current technologies, for example, CMOS and silicon, to SiGe, silicon-on-insulator (SOI), carbon nanotube based interconnects, biochip, molecular electronics or other approaches designed to give greater performance and/or reduce power requirements.
Concurrently, the BE-wafer is created to form its metalized layers 22404 (
The FE-wafer and BE-wafer are then aligned relative to each other (
Yet another alternative approach is illustrated in
At this point it should be noted that a further advantage to these approaches is that, if necessary, some further rerouting of connections can be made on the FE-wafer or BE-wafer (or possibly both). As a result, it is even possible to create the FE- and BE-wafers to be more generic, with the other providing suitable connection locations for a particular application. Moreover, at this point, the combined FE/BE-wafer or FE/BE/(FE-wafer or chip) stack can be treated like any other wafer created using wholly conventional processes, and thus can be a mother or daughter wafer with respect to other wafer(s) for purposes of the subject matter described herein.
Still further, chip units can be designed that use much higher speed communication between chips than are available with wired connections, due to problems related to cross-talk causing interference, through use of chip-to-chip optical connections. For example, by placing a semiconductor laser on one chip in a stack and a corresponding photo detector on the other chip in the stack to which it is mated, an optical—rather than wired—connection can made between the two. If the two are sufficiently close to each other, the possibility of even optical crosstalk is minimized. This aspect is illustrated in simplified fashion in
Detailed Contact and Material Alternatives
As will now be appreciated, the contacts fairly complex aspects, in and of themselves due to the nature of the tack and fuse processes, reiterated in simplified form in
In general, whatever the application, a daughter wafer contact 23802 of
Referring back to
Barrier Layer: Ti/W+Pd
Standoff Layer: Absent
Diffusion/Malleable Layer: Gold/Tin (80/20) (between 1 and 12 microns)
Cap/Adhesion: Gold (>500 Angstroms; Typically 1500 to 10,000 Angstroms)
Oxidation Barrier: the Cap/Adhesion layer serves as this layer also. Note, while the malleable layer may be composed of any combination of standoff, diffusion, cap and barrier layer, here the malleable is the combination of the diffusion and cap layers.
Similarly, for the mother contact (with reference to
Barrier Layer: Absent for Cu/Al pads
Rigid: Copper (>2 microns)
Diffusion Barrier Layer: Nickel (5000 Angstroms; typically 0.5 to 3 microns)
Cap/Diffusion: Gold (>500 Angstroms; typically 1500 to 10,000 Angstroms)
With respect to the above, the following sets forth further, non-exhaustive, alternative materials that can be used for the specified contact layers.
Barrier (mother or daughter)/Diffusion Barrier (mother): This could be, for example, Ni, Cr, Ti/Pt, Ti/Pd/Pt, Ti/Pt/Au, Ti/Pd, Ti/Pd/Au, Ti/Pd/Pt/Au, TiW, Ta, TaN, Ti, TaW, W, or could be absent if the IC pad is made of the same material as the standoff layer. Standoff Layer (daughter)/Rigid Layer (mother): Ni (especially if barrier is Ni), Cu (especially if pad is Cu), Al, Au, W, Pt, Pd, Co, or Cr. If sputtered rather than plated, then any type of metal which has a melting temperature higher (typically >50° C. higher) than the melting temperature of the malleable (diffusion) material. It could also be made of any of the barrier materials.
Malleable (Diffusion) Material: A metal that melts at low temperature like: tin, indium, lead, bismuth, aluminum, zinc, magnesium or other material with melting point less than 1000° C. or an alloy combining two or more of those together, or combining one or more of those together with a higher temperature melting material like gold, silver, copper, titanium, or other analogous material. Combination examples include: Au/Sn, Cu/Sn, Cu/Zn, Bi/Ag, etc. Note: An important aspect for this selection is that it is not desirable for the selected material to actually melt during the attach process since that would be too slow of a process, adding to cost, and could cause problems with creep or running causing contact shorting and thus limiting the density. It is the malleable/rigid combination which ultimately gives the strength of the contact. Typically an alloy containing compounds with mixtures of one or more of: Au, Ag, Bi, Cd, Cu, Fe, In, Pb, Sn, Sb, or Zn are good choices. The primary condition is that the melting temperature should be less than or equal to the melting temperature of the rigid post and, if present, the standoff layer. Typically, the malleable should have a melting point of at least 50° C. lower than the melting point of the rigid, although we have used a melting point differential of between 100° C. to 500° C. Advantageously, the malleable material can also be built up of several materials to give the proper the height needed to overcome non-planarity of the contacts. In fact the malleable material can be built on top of a standoff post of the rigid material. For example, in one case, the malleable material could consist of Au/Sn, 5 microns high. Alternatively, in another case, the post could consist of a stack of a rigid material such as 4 microns of nickel covered by a thinner layer, for example, 1 to 1.5 microns, of the malleable material.
Malleable Cover Material (Cap/Adhesion layer): These can be a material that could become wet under temperature, such as a low-temperature metal (or alloy) like tin, indium, lead or zinc. Note that this cover material layer is generally much thinner than the malleable material layer. For example, it would normally be around 10 to 20 times thinner. For example, if the malleable (plus any standoff) material were 5 microns high, the malleable cover material could be 0.5 microns, and would typically in the range of 0.1 micron to 1 micron (or about 50× to 5× thinner than the malleable layer). One good example of such a cover is tin (Sn). Such a cover material will have a low melting temperature and can turn liquidus at the tack temperature. However, because the layer is very thin, it will not cause shorting between adjacent contacts as there is not enough liquid; to do so. At the same time it can make for a quicker attach process to the rigid cap, because the tack phase becomes a liquid process. In general, this cover should be selected so as to be compatible with the malleable material so that, after fuse, the resultant combination would be suitable for a strong bond. For the tin example, such an approach would typically use a Au/Sn contact with a Sn cap.
Malleable Cover Material (Oxidation Barrier)/Rigid Cover Material (diffusion cap): If the adhesion layer is used for the “tack” process and it is a material prone to oxidize, like tin or zinc, then it should be covered with a very thin oxidation barrier. Otherwise, a reactive gas or liquid should be used during the tack process to remove the oxide, or a high enough pressure must be used to break through the oxide, as can happen for example, if indium is used as the cap. The cover can even be an epoxy. For most materials, a thickness of 10 times thinner than the cap itself will work. Note again that the malleable cover could be of a higher temperature material that only becomes a lower-temperature alloy (or only becomes a bonding agent) when the malleable cover material comes into contact with and begins to mix with the rigid cover material or with the malleable material. For example if the two covers were two parts of a mixable epoxy or if the oxidation barrier were gold and the malleable gold-tin then the intermixing of the tin into the oxidation layer during the attach process would cause that material to have a lower melting point. In general, this layer can be any metal/material which does not readily oxidize (e.g. Au, Pt, etc.).
Connection-Related Tooling
Having described numerous different approaches for interconnection of chips on a chip, die and wafer basis, and various details that make it possible to employ many permutations, variations and combinations thereof, it is useful to diverge and describe certain different types of tooling that have been devised and can be advantageously used to assist in the joining process. Note that none of these tooling approaches are essential for accomplishing any of the permutations, variations or combinations, but rather they have been developed to ease the process and can be used for other chip-related operations, like “pick and place,” particularly where it is desirable to simultaneously do so for multiple chips at the same time and even more advantageously, in circumstances where those chips vary in height with respect to each other.
For purposes of explanation, different tooling variants will be described with respect to use in the tack and fuse process, since an understanding of that approach will obviate the need to describe the simpler uses, since they will be a subset or trivial variant thereof.
As described herein, the attachment process is split into two parts: A first part in which chips are lightly attached together (the “tack” phase) and a second part, the “fuse” phase that provides the bond strength. The tack process heats up the contacts and keeps them abutting under light pressure to allow the materials on the two corresponding contacts to interdiffuse into one another.
During this process, if the force of gravity alone is not sufficient to provide for the requisite pressure, a small amount of pressure can be applied to ensure that the chips do not move during the process reducing the prospect of mechanical shock or non-uniformity in attachment, either of which could result in less than adequate adhesion between the contacts and thereby an inability to withstand wafer handling. In addition, the pressure can help ensure that if any local heating causes the malleable material to become partially or completely liquidus (or simply become more malleable than ideal without becoming liquidus, and counteract the pressures and surface tensions or other forces that might otherwise push the pieces apart or, in the case where excess softness of the malleable material occurs, it can prevent excess lateral movement of the parts individually and collectively. Thus, the application of slight pressure can ensure more latitude in the temperatures and handling conditions for the fuse process to account for manufacturing tolerances and variations.
However, one of the problems with putting pressure on these chips is that if the base element, for example a wafer, has multiple chips which are attached to it, the individual chips might not be co-planar and could even differ significantly in height. Thus, if one were to simply place a flat surface or plate on the top of the chips, the applied pressure would be unevenly applied.
As illustrated below, the approaches devised for addressing the foregoing is to use an arrangement between the source of the force and the chips that will conform to or account for the different heights and thus allow all chips to have equal pressure applied to them.
One approach for accomplishing this uses a series of pins or posts that match with the individual chips on a one-to-one basis as the arrangement. Two different variants of this approach will now be described with the understanding that other variants can be devised by, for example, combining aspects from each or from the other tooling approaches as described below.
As illustrated in
Depending upon the particular implementation, the face of any particular pin or post can be: flat, an inverse die of the chip it will apply pressure to, or some other shape appropriate for the particular application. In addition, the pin or post itself at or near the face (as well as along some or all of its length) can have a circular cross section or some other, non-circular (i.e. oval, quadrilateral, hexagonal, octagonal, etc.) closed shape. Moreover, the perimeter and planar area of the face can be larger or smaller than the perimeter or area of the particular chip it will contact (i.e. it can extend beyond the periphery of the chip or be wholly or partially contained within it, the important aspect being that the face is configured to apply force to the chip without damaging it, particularly without cracking or chipping it.
In use, the posts within the frame (and in some cases, the frame itself) are brought downward, in an unconstrained condition, until each post is in appropriate contact with its respective chip (
Thereafter, the joining process can continue as described herein or in some other manner.
Note that the individual pins/posts or groups (if multiple pins/posts per chip) need to be wide enough to ensure that any pressure transferred by them to the chips does not crack the chips and they should be placed so that they do not clip an edge or a corner of a chip during the process.
In both cases, by using the frame to hold the posts or pins, once constrained, the posts or pins can only meaningfully move in the vertical direction, allowing the structure to only apply vertical pressure while conforming to the topography of the chips attached to the wafer.
Advantageously, as noted herein, where a tack and fuse approach is used, the forces needed for the “tack” step will typically be on the order of 1 gram per contact or less and for the fuse process, typically less than 0.001 grams per contact. As a result, the pins or posts can be readily constrained within the frame without difficulty through, for example a clamping or other locking approach, the particular approach being a matter of design choice and unimportant for understanding the tooling and its use.
Advantageously, in some implementations, either of the above tooling can be further enhanced by making it possible to apply a vacuum to the chips. In the case of the pin/post-per-chip tooling, this can be accomplished by providing passageways 24412, 24414 through the post and openings on the post face 24406. Alternatively, with the group-of-pins/posts approach, the pins/posts themselves can house the passage through which the vacuum is drawn. Alternatively, by selecting the appropriate shapes and spacings for the pins/posts, passages among abutting pins can be formed (within the chip boundary) or eliminated (near the chip periphery) so as to allow for the vacuum to be drawn through those interstitial passages.
In either tooling instance, with such a variant, a vacuum can be applied to the chips to, for example, allow for the tooling itself to be used in a pick-and-place operation or for the vacuum to further inhibit non-vertical (i.e. undesirable) movement of the chip during, for example, the tack or fuse processes.
Through a further alternative approach, a material can be applied to the face 24406, 24606 of the pins or posts which will cause them to initially adhere to the chips, but is also selected so as to be able to “detach” from the chip when the operation is complete. For example, a material can be used on the faces that will liquefy and run, melt or vaporize at about the tack or fuse temperature but, in doing so will not damage the chips and, if it leaves a residue on the chip or element to which the chip is attached, the residue can be removed through some non-damaging process or ignored without detrimental effect.
While the post/pin solution provides only vertical motion, some implementations of that approach do not actually hold the chips in place and, in some cases, cannot guarantee that force will be uniformly applied across each chip or that the chips will not tilt in angle during, for example, the tack or fuse process. Thus, in some cases, movement of the chips or non-uniform fusing, across individual chips or between chips with different heights, could occur.
In such cases, an alternative tooling approach, shown in
Moreover, depending upon the particular material used between the plate and chips, the material can be reusable for two or more cycles of pressure application and joining or it can strictly be a one-time use material.
As with the pin/post variants, and shown in
Alternatively, and advantageously, this arrangement can also be used along with pin/post based tooling if the particular application renders it less desirable to apply the force to the pins/posts via the frame. In such an arrangement, the pin based tooling is applied as above. However, if the pins/posts are all of equal height, once brought into contact with the chips, the ends of the pins/posts will reflect the same height differentials as the chips. However, by using the plate and material arrangement on the ends of the pin/posts opposite the chips, the height differential can be accommodated and the appropriate force easily and uniformly applied. Moreover, through this approach, the particular material will likely be sufficiently removed physically from the chips that it need not be a temperature resistant as those materials that must be brought into direct contact with the chips.
Another alternative approach to maintaining the chips in contact with the element(s) to which they will be joined, that is similar to the plate variant of
This body 25000 is then placed on the array of chips 24906 so that the hardenable material 25004 adheres to each while being maintained in a level position (
Once hardened, the chips can be moved to the element to which they will be attached, and the body can be weighted with a separate and removable weight, if necessary, during the attachment process (if needed) (
Optionally, an underfill 25302 material can be flowed in between the body and the element to which the chips will be attached (
Once joined, and after removing the weight if a weight was used or applying the underfill (if done), the entire (or a large portion of the) body can be removed (
Similarly, this “body” approach can be used in conjunction with a pin/post based tooling to account for differences in pin/post heights and allow for application of the force through other than direct application to the frame. In such a case, the pins/posts are brought into contact with the chips, the body is then brought into contact with the ends of the pins/posts opposite the chips and hardened. Thereafter the force is applied as above in the desired process. Once the chips are attached, the pin/post-frame-overall body combination can be readily removed from the chips as with the normal pin/post approaches. Thereafter, the overall body can be separated from the pin/post-frame tooling through any convenient process that would soften or remove the hardenable material or by simply cutting or shearing off the pins at a point outside the hardenable material.
Moreover, a further advantage to this particular combination approach is that it allows for repeatability in cases where an assembly-line approach to joining a multiple chips to one or more respective underlying elements and, as noted above with respect to certain variants, use as part of a pick-and-place approach.
Finally, with respect to all of the above tooling as well as other variants, permutations or combinations thereof, it should be noted that, if required for a particular use, a gas, like forming gas or formic acid or a flux can be flowed in between the frame and the chip during the fuse portion of the process.
Note that, in some cases, the pin/post approach will be preferable to use of some flexible or spongy materials (i.e. those which could themselves apply too much lateral pressure on the chips, causing them to tilt or to shift during the fuse process, or could require extremely (and commercially impractical) tight tolerances with respect to the fuse process conditions)).
In summary reiteration, although the invention has been described in connection with particular types of chips including optical chips (i.e. ones carrying, for example, one or more lasers, one or more photodetectors, or some combination thereof) however, the approaches described herein can be used equally well to create “through-chip” electrical connections in any kind of doped semiconductor chip comprising transistors or other electronic circuit components in addition to, or instead of, optical components.
Similarly, although certain materials have been identified as suitable for use as “post and penetration” contact materials, those materials should not be considered literally the only materials that can be used, since the important aspect is the relative hardness between the two such that diffusion between the two occurs to form the connection, not the particular materials used. Since the particular pairings of materials will, to some extent, be determined by factors such as availability, cost, compatibility with the other components being used or other manufacturing-related processes that are unrelated to those described herein, it is unhelpful to itemize more than a few of the potentially limitless pairs of materials. Similarly, there are a number of optically transmissive materials beyond optical epoxies. However, the criteria for selection of the particular material that would be used for a particular application may be affected or governed by other factors not pertinent to the subject matter herein. Accordingly, it should be understood that any optically transmissive medium (or media) that could be inserted into the void and transmit laser light as required for the particular application should be considered as being a suitably usable material without specific itemization of all possible alternatives thereof.
It should thus be understood that this description (including the figures) is only representative of some illustrative embodiments. For the convenience of the reader, the above description has focused on a representative sample of all possible embodiments, a sample that teaches the principles of the invention. The description has not attempted to exhaustively enumerate all possible variations. That alternate embodiments may not have been presented for a specific portion of the invention, or that further undescribed alternate embodiments may be available for a portion, is not to be considered a disclaimer of those alternate embodiments. One of ordinary skill will appreciate that many of those undescribed embodiments incorporate the same principles of the invention and others are equivalent.
This application is a Division of U.S. patent application Ser. No. 11/556,747, filed Nov. 6, 2006, which is a Continuation of U.S. patent application Ser. No. 11/422,551, filed Jun. 6, 2006, which claims priority to U.S. Provisional Patent Application No. 60/690,759, filed Jun. 14, 2005. The disclosures of each of the foregoing applications are incorporated herein by reference in their entireties as if fully set forth herein.
| Number | Date | Country | |
|---|---|---|---|
| 60690759 | Jun 2005 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 11556747 | Nov 2006 | US |
| Child | 12846495 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 11422551 | Jun 2006 | US |
| Child | 11556747 | US |
