Interlocking conductor method for bonding wafers to produce stacked integrated circuits

FIELD OF THE INVENTION

0001[0001] The present invention relates to integrated circuits, and more particularly, to a method for bonding wafers together to form integrated circuits having a stack of thin layers.

BACKGROUND OF THE INVENTION

0002[0002] Modem integrated circuits are typically constructed in a thin layer in a semiconducting layer on a substrate wafer such as silicon. This essentially two-dimensional structure limits both the size of the integrated circuit and the speed at which the circuit operates. The speed at which an integrated circuit operates is determined by the distance between the farthest separated components that must communicate with one another on the chip. For any given number of components, the path lengths will, in general, be significantly reduced if the circuit can be laid out as a three dimensional structure consisting of a number of vertically-stacked layers of circuitry, provided the vertical distances between the layers are much smaller than the width of the chips that make up the individual layers.

0003[0003] One promising scheme for providing such stacked structures utilizes a method for stacking and bonding entire wafers. In this method, integrated circuits are fabricated on conventional wafers. Two wafers are bonded vertically by thinning one wafer in a first coarse thinning operation by removing material from the back of the wafer. The circuitry on the front surface of each wafer is covered with an insulating layer having metal filled vias that make contact with the underlying circuitry and act as electrical connection points between the two wafers. In addition, the front surfaces of the wafers include bonding pads that are planar metal areas that do not connect to the underlying circuitry. The bonding pads are provided to increase the bonded area. The front surfaces of the wafers are then placed in contact with one another so that the bonding pads on one wafer are in contact with the pads on the other wafer. Thermal diffusion bonding is then used to bond the metal pads, and hence, the wafers together. One of the wafers is then further thinned to a thickness of a few microns by etching or mechanically grinding the back surface of that wafer further. Once the wafer has been thinned, a new set of vias is opened in the backside and filled with metal to provide the connection points to the pads on the front side of the wafer that make connections with the circuitry in the wafer. In addition, a new set of bonding pads is formed on the backside of the wafer so that another wafer can be bonded to the stack. The process is then repeated until the desired number of layers has been bonded to form the three-dimensional stack. The three-dimensional stack is then cut into three-dimensional chips and packaged.

0004[0004] Each time a top wafer in the stack is thinned, the wafers below that wafer are subjected to a significant amount of lateral stress by the grinding process. Accordingly, the bonds that hold the wafers together must withstand these large stresses. In principle, the ability of the wafers to withstand these stresses can be increased by increasing the bond area, i.e., devoting more area to bonding pads between the wafers. Unfortunately, this approach has the drawback of reducing the number of vertical conductors that can be provided in circuits requiring a high density of inter-layer vertical conductors.

0005[0005] This process also requires a considerable amount of “backside processing” to provide the new set of contacts on the backside of the wafer. The generation of the vertical conductors requires a number of masking and deposition steps. Each conductor is structurally similar to a golf tee. The “head” of the tee provides the area needed to accommodate alignment errors. Separate mask and deposition steps are required to etch the via that forms the stem of the tee and the head that sits on this stem. The via must also be lined to prevent diffusion of the metal into the surrounding silicon. Such extensive backside processing can substantially reduce the yield of stacked wafers.

0006[0006] The situation is made more complicated by the lack of fiduciary marks on the backside of the thinned wafer. Hence, precise alignment of the masks that define the locations of the vias with respect to the circuitry on the front side of the wafer is difficult. Misalignment of these vias leads to defects that render the entire stack of chips useless. In addition, the semiconductor processing steps required by the backside processing of the top layer subject the underlying layers to thermal and mechanical stresses that are repeated with each new layer.

0007[0007] Broadly, it is the object of the present invention to provide an improved method for stacking and thinning wafers to generate a three-dimensional integrated circuit.

0008[0008] It is a further object of the present invention to provide a method that eliminates the backside processing steps utilized in prior art stacking techniques.

0009[0009] These and other objects of the present invention will become apparent to those skilled in the art from the following detailed description of the invention and the accompanying drawings.

SUMMARY OF THE INVENTION

0010[0010] An integrated circuit wafer element and an improved method for bonding the same to produce a stacked integrated circuit. An integrated circuit wafer according to the present invention includes a substrate having first and second surfaces constructed from a wafer material, the first surface having a circuit layer that includes integrated circuit elements constructed thereon. A plurality of vias extend from the first surface through the circuit layer and terminate in the substrate at a first distance from the first surface. The vias include a stop layer located in the bottom of each via constructed from a stop material that is more resistant to chemical/mechanical polishing (CMP) than the wafer material. The vias may be filled with an electrically conducting material to provide vertical connections between the various circuit layers in a stacked integrated circuit. In this case, the electrical conducting vias may also be connected to various circuit elements by metallic conductors disposed in a dielectric layer that covers the circuit layer. A plurality of bonding pads are provided on one surface of the integrated circuit wafer. These pads may be part of the conducting plugs resulting from filling the vias with the conducting material. These pads preferably extend above the surface of the integrated circuit wafer. A stacked integrated circuit according to the present invention is constructed by bonding two integrated circuit wafers together utilizing the bonding pads. One of the integrated circuit wafers is then thinned to a predetermined thickness determined by the depth of the vias by chemical/mechanical polishing (CMP) of the surface of that integrated circuit wafer that is not bonded to the other integrated circuit wafer, the stop layer in the vias preventing the CMP from removing wafer material that is within the first distance from the first surface of the substrate of the wafer being thinned. The thinning operation leaves a portion of the plug in the via standing above the surface of the thinned substrate. A third wafer can then be bonded to the raised portion of the plug by bonding the raised portion of the plug to pads on the third wafer's surface. The bonding pads on the third wafer surface preferably include a depressed region for engaging the raised portion of the plugs on the second wafer.

BRIEF DESCRIPTION OF THE DRAWINGS

0011[0011]FIG. 1 is a cross-sectional view of a portion of a stacked integrated circuit <b>10</b> according to the present invention.

0012[0012]FIG. 2 is a cross-sectional view of a wafer <b>100</b> used as a starting point for a component layer.

0013[0013]FIG. 3 is a cross-sectional view of wafer <b>100</b> after a via <b>120</b> has been etched through the dielectric layers and into substrate <b>110</b>.

0014[0014]FIG. 4 is a cross-sectional view of wafer <b>100</b> after via <b>120</b> has been lined with two layers.

0015[0015]FIG. 5 is a cross-sectional view of wafer <b>100</b> after a trench <b>128</b> has been etched in dielectric layer <b>116</b>.

0016[0016]FIG. 6 illustrates a copper pad that is flush with the surrounding dielectric after the excess copper has been removed by chemical mechanical polishing (CMP).

0017[0017]FIG. 7 is a cross-sectional view of a completed component layer element <b>135</b>.

0018[0018]FIG. 8 is a cross-sectional view of a base component layer element <b>202</b> positioned relative to a component layer element <b>201</b> that is to be bonded to element <b>202</b>.

0019[0019]FIG. 9 is a cross-sectional view of the component layers after component layer element <b>202</b> has been thinned.

0020[0020]FIG. 10 is a cross-sectional view of component layer <b>135</b> shown in FIG. 7 after a socket <b>282</b> has been formed from a SiO₂layer <b>281</b>.

0021[0021]FIG. 11 is a cross-sectional view of component layer <b>135</b> shown in FIG. 7 after socket <b>261</b> has been etched.

0022[0022]FIG. 12 is a cross-sectional view of a portion of a component wafer.

DETAILED DESCRIPTION OF THE INVENTION

0023[0023] The manner in which the present invention provides its advantages may be more easily understood with reference to FIG. 1 which is a cross-sectional view of a portion of a stacked integrated circuit <b>10</b> according to the present invention having a base component layer <b>20</b> and two stacked component layers shown at <b>30</b> and <b>40</b>. Each component layer includes an integrated circuit layer that is constructed on a substrate using conventional integrated circuit fabrication techniques. To simplify the following discussion, it will be assumed that the integrated circuit layer is constructed on a conventional silicon substrate in the form of a wafer. The integrated circuit layers corresponding to component layers <b>20</b>, <b>30</b>, and <b>40</b> are shown at <b>22</b>, <b>32</b>, and <b>42</b>, respectively. The substrates on which these layers were constructed are shown at <b>21</b>, <b>31</b>, and <b>41</b>, respectively. The integrated circuit layer is covered with one or more layers of dielectric such as SiO₂in which various metal conductors are constructed and connected to the circuitry by vias. To simplify the drawing, only the metal conductors that are to be connected to components on other component layers are shown in the drawing. Exemplary conductors of this type are shown at <b>25</b>, <b>35</b>, and <b>45</b> together with the dielectric layers that are shown at <b>23</b>, <b>33</b>, and <b>43</b>.

0024[0024] Connections between the various component layers are provided by vertical conductors that pass through one or more component layers. A typical vertical conductor is shown at <b>50</b>. Vertical conductor <b>50</b> is constructed from component conductors shown at <b>51</b>-<b>53</b> by thermal diffusion bonding of the component conductors. The thermal diffusion bonding of the component conductors also bonds the various component layers together.

0025[0025] It should be noted that, in general, there are thousands, if not tens of thousands, of vertical conductors in a typical stacked integrated circuit. Hence, the diameters of the vias are preferably as small as possible. The minimum diameter of a via is determined by the aspect ratio permitted by the metallization process used to fill the via. Vias with aspect ratios of greater than <b>5</b> are difficult to fill reliability. Hence, it is advantageous to have the component layers be as thin as possible. In addition, thin component layers are more flexible. The flexibility improves the strength of the stacked structure and reduces cracking or other damage caused by thermal stress.

0026[0026] It should also be noted that it is important that the component layers be planar sheets having parallel top and bottom edges. In general, a stacked integrated circuit according to the present invention is constructed by bonding wafer-sized component layers. After all of the layers have bonded, the stacked structure is then divided into individual stacked chips. If the component layers become wedge shaped or have hills and valleys in the surface thereof due to fabrication errors, the bonding between layers will fail. In addition, the vertical vias will not be properly aligned in some areas of the chip. Hence, any economically practical wafer-stacking scheme must assure a high degree of precision over the entire wafer for each wafer component used. The manner in which the present invention provides this high degree of precision will now be discussed in detail.

0027[0027] For the purposes of the present discussion, the side of a wafer that contains the integrated circuits and conventional metallization layers such as the layers that contain conductors <b>25</b>, <b>35</b>, and <b>45</b> will be referred to as the “front side” of the wafer. The opposite side will be referred to as the “backside”. A stacked wafer circuit such as that shown in FIG. 1 is constructed by first bonding wafers <b>20</b> and <b>30</b> via their front sides, and then bonding successive wafers such as wafer <b>40</b> by bonding the front side of the new wafer to the backside of the last wafer in the stack.

0028[0028] Refer now to FIG. 2, which is a cross-sectional view of a wafer <b>100</b> used as a starting point for a component layer. It will be assumed that wafer <b>100</b> has its active circuit layer <b>112</b>, which is covered with a dielectric layer <b>113</b>, in place. As noted above, various metal conductors are typically constructed in the dielectric layer and connected to the circuitry by metal filled vias. Typical metal conductors are shown at <b>114</b> and <b>115</b>. These conductors can be divided into two classes, those that provide connections between the various components in integrated circuit layer <b>112</b> and those that are to provide connections to components in other layers of the final stacked integrated circuit. Conductor <b>115</b> is in the first class, and conductor <b>114</b> is in the second class. It will also be assumed that a second layer of dielectric <b>116</b> covers the conductors.

0029[0029] Refer now to FIG. 3, which is a cross-sectional view of wafer <b>100</b> after a via <b>120</b> has been etched through the dielectric layers and into substrate <b>110</b>. As will be explained in more detail below, the depth <b>121</b> by which via <b>120</b> extends into substrate <b>110</b> is critical. Preferably, via <b>120</b> is etched in two steps. In the first step, the via is etched using an etchant that stops on the silicon substrate such as a fluorocarbon-based plasma etch. In the second step, the via is extended into substrate <b>110</b> by 4 to 9 microns using a timed halogen-containing gaseous plasma. It should be noted that the placement of the vias can be controlled precisely, since the wafer has fiduciary marks that are visible from the front side of the wafer, and these marks can be used to align the masks that define the via locations using conventional alignment tools.

0030[0030] Refer now to FIG. 4, which is a cross-sectional view of wafer <b>100</b> after via <b>120</b> has been lined with two layers. Layer <b>125</b> consists of a thin dielectric layer, preferably 0.05 to 0.10 microns of SiO₂. This layer acts as an electrical insulator to prevent shorting between the metal layer of the filled via and components in the integrated circuit layer <b>112</b>. The second layer <b>126</b> consists of a thin layer of SiN, typically 0.05 to 0.10 microns in thickness. The SiN layer serves two functions. First, it provides a diffusion barrier that helps to prevent the metal used to fill via <b>120</b> from diffusing into the integrated circuit layer if the primary diffusion barrier discussed below fails. Second, the silicon nitride provides an etch stop for the chemical etching processes used in the thinning of the silicon wafer. For example, the silicon can be thinned using a wet chemical process such as a substituted ammonium hydroxide or other alkaline chemical etches. It should also be noted that this etch stop will provide some resistance to acidic etch solutions. In this case, the silicon nitride acts as the etch stop. If a dry etch such as a Cl₂based plasma chemistry is used to thin silicon, the SiO₂layer can be used as an etch stop.

0031[0031] Refer now to FIG. 5, which is a cross-sectional view of wafer <b>100</b> after a trench <b>128</b> has been etched in dielectric layer <b>116</b>. A via <b>129</b> is opened in the bottom of trench <b>128</b> to provide contact with pad <b>117</b> that provides electrical connection to components in circuit layer <b>112</b> that are to be connected to the vertical conductor that will be formed by filling via <b>120</b> with metal. A third layer <b>130</b> is deposited in via <b>129</b> and trench <b>128</b>. Layer <b>130</b> serves two functions. First, layer <b>130</b> acts as a diffusion barrier that prevents the metal used to filled the via and trench from diffusing into the remainder of the wafer. In the preferred embodiment of the present invention, the preferred metal is copper. The diffusion barrier is preferably Ta, TaN, or WN or other ternary barrier material such as Ta_xSi_yN_z, W₂Si_yN_z, etc. A 200-1000 A° barrier layer is preferably deposited by a CVD or a PVD process such as sputtering. Second, the portion of layer <b>130</b> at the bottom of via <b>127</b> acts as a stop in the wafer thinning process described below. Trench <b>128</b> is then filled with metal.

0032[0032] The preferred metal for the filling operation is copper. In embodiments utilizing copper, a copper seed layer is deposited in the trench and vias prior to the deposition of the copper. The seed layer can be deposited utilizing a CVD or a sputtering process. The seed layer maintains the proper conduction during the subsequent electro-plating process utilized to deposit the metallic copper. After the seed layer is deposited, the trench is filled with copper using electrochemical plating. The excess copper is removed by chemical mechanical polishing (CMP), leaving a copper pad that is flush with the surrounding dielectric as shown in FIG. 6. In the preferred embodiment of the present invention, the final copper pad <b>132</b> is elevated relative to the surrounding dielectric layer <b>133</b> by 0.01-0.2 microns as shown in FIG. 7, which is a cross-sectional view of a completed component layer element <b>135</b>. This slightly elevated pad provides improved bonding when the component layer element is bonded as described below. The elevation of the pad can be accomplished by lowering the surrounding dielectric layer or by increasing the height of the copper. The dielectric layer can be lowered by selective etching using a fluorine containing etch process. The copper height can be increased by electroless deposition of additional copper, which will occur only on the exposed copper surface.

0033[0033] The manner in which the first component layer is added to the base component layer by the front side bonding of two wafers will now be explained in more detail with reference to FIG. 8. FIG. 8 is a cross-sectional view of a base component layer element <b>202</b> positioned relative to a component layer element <b>201</b> that is to be bonded to element <b>202</b>. The elements are positioned by turning element <b>201</b> over such that its copper bonding pads are positioned over the corresponding bonding pads on element <b>202</b>. To simplify the drawing, only one pair of pads is shown at <b>210</b> and <b>211</b>; however, it is to be understood that each component element may have thousands or even millions of such pads. The two component elements are pressed together and bonded using thermal diffusion bonding. The wafers are bonded by compressing the two wafers using 20-60 psi pressure at 300-450° C. temperature in a nitrogen or air atmosphere for 5-50 minutes. The wafers are positioned by utilizing fiducial marks on the front sides of the wafers. The marks on the front side of wafer <b>202</b> are viewed from the backside of the wafer. To improve the accuracy of the alignment, wafer <b>202</b> may be thinned prior to bonding.

0034[0034] After the two elements have bonded, element <b>202</b> is thinned further to a thickness of a few microns as shown in FIG. 9 which is a cross-sectional view of the component layers after component layer element <b>202</b> has been thinned. As noted above, the resulting layer component must have parallel surfaces to assure that any subsequent element bonded to this element will be properly aligned and bonded. In the preferred embodiment of the present invention, the portion of the diffusion/stop layer shown at <b>130</b> in FIG. 5 in the bottom of the vertical vias is used as a stop for this thinning process. The preferred thinning process utilizes CMP of the substrate <b>203</b>. The thinning process can be a combination of grinding and CMP and/or etch processes. For example, a CMP process will remove the silicon substrate at a rate that is 100 times faster than Ta in layer <b>130</b>. Hence, the CMP process will stop at the same point on each of the vias. The depth of the vias, as noted above, can be controlled to a high degree of precision. Hence, the resulting component layer will have a thickness that is tightly controlled, since it is determined by the depth of the vias.

0035[0035] The silicon substrate is etched faster than the metal filled vias; hence, when the thinning process is completed, the end of the metal via <b>252</b> is elevated relative to the surface <b>253</b> of substrate <b>203</b>. At this point, no further processing of the backside of wafer <b>202</b> is required.

0036[0036] The stacking process can be continued by connecting pads on the front side of another wafer to the elevated posts on the backside of the last wafer that was added to the stack. In the preferred embodiment of the present invention, a “socket” is provided on the front side of each new wafer. The sockets engage the elevated “posts” described above. The sockets can be generated by any of a number of methods. One such method is illustrated in FIG. 10, which is a cross-sectional view of component layer <b>135</b> shown in FIG. 7 after a socket <b>282</b> has been formed from a SiO₂layer <b>281</b>. The thin layer of SiO₂can be deposited over the front side of the next wafer by conventional techniques and a via opened in the layer to form the socket. The thickness of the SiO₂layer must be less than the height with which conductor <b>252</b> extends over substrate <b>253</b>.

0037[0037] The depression that forms the socket can also be generated by etching a depression <b>261</b> in the metal pad <b>262</b> on the front side of the component wafer as shown in FIG. 11, which is a cross-sectional view of component layer <b>135</b> shown in FIG. 7 after socket <b>261</b> has been etched. The etch area can be set by conventional masking operations. Once again, the depth of the depression should be less than the height with which conductor <b>252</b> extends over substrate <b>253</b>.

0038[0038] Once the sockets have been formed in the front surface of the next component wafer to be added to the stack, the wafer is positioned over the backside of the last wafer in the stack and bonded thereto as described above. The positioning operation preferably utilizes a new front-side fiducial mark that is generated using the filled vias as references. Alternatively, the filled vias can be used as fiducial marks.

0039[0039] After bonding, the new component element is thinned as described above. This process may be continued with additional component elements until the desired stack thickness is obtained.

0040[0040] The above-described embodiments of the present invention utilize a socket that mates with the extended post on the backside of the last component layer in the stack. The socket provides additional resistance to lateral stress such as that generated in the thinning operations. However, if there are a sufficient number of binding sites between the component wafers, this additional stress resistance is not needed. In such cases, the extended posts can be bonded directly to the metal pads on the front side of the next wafer to be added.

0041[0041] The above-described embodiments of the present invention have utilized vertical conductors that serve both as bonding points between the component wafers and also as signal or power conduction paths that run vertically through the stack. In addition, there must be a sufficient density of posts in the wafer to provide the etch stop function during the thinning process. If the density of posts is too low, “dishing” can occur in the regions between the posts. In the preferred embodiment of the present invention, the distance between vias is less than 50 μM and the wafers are thinned to a thickness of 1 to 25 μM.

0042[0042] If the particular integrated circuit design does not provide sufficient vertical conductors to provide enough bonding points to assure that the wafers remain bonded together, additional bonding points can be provided. Such bonding points are preferably similar to those used for vertical conduction paths. However, since these bonding pads do not need to connect to the internal conductors in the component wafers, the lateral extent of such pads may be substantially less. A typical bonding point is shown at <b>280</b> in FIG. 12, which is a cross-sectional view of a portion of a component wafer.

0043[0043] It should be noted that the bonding points that hold the various component wafers together extend through the component wafers. This arrangement provides a significant improvement in the stress levels that can be accommodated by the bonded wafers. For example, prior art schemes that utilize metal pads that are deposited on the opposing surfaces of the bonded layers provide less resistance to lateral stresses. To provide a resistance similar to that provided by the present invention, the area of the pads must be substantially increased. However, this can led to “peeling” of the pads off of the surface of the wafers because of the different thermal expansion coefficients of the pads and the underlying silicon.

0044[0044] Various modifications to the present invention will become apparent to those skilled in the art from the foregoing description and accompanying drawings. Accordingly, the present invention is to be limited solely by the scope of the following claims.