The present disclosure relates to a semiconductor package, and more particularly, to a semiconductor structure having a die-stacking structure.
As the artificial intellectual (AI) models have been applied to more and more fields, a demand for suitable hardware having greater computation capability also raises. Since the AI models usually require large amounts of parallel computing, most of the computation hardware includes multiple cores, which requires large circuit area. Therefore, researches have been conducted to embed more computation power within a smaller area. However, as the computing power per unit area gradually increases, heat dissipation also becomes a major challenge in packaging design. In addition, as the computing power continues to improve, the bottleneck of system performance may gradually shift towards the memory bandwidth or the data sharing speed between different cores, which make the packaging design even more complicated. Therefore, providing a semiconductor structure that can facilitate greater computation capability within a smaller area has become an issue to be solved.
One aspect of the present disclosure provides a semiconductor package. The semiconductor package includes a first dielectric layer, a first redistribution layer (RDL), a first bonding layer, a plurality of bottom dies, a second dielectric layer, a plurality of conductive pillars, a second bonding layer, a plurality of top dies, a third dielectric layer, and a plurality of solder bumps. The first RDL is disposed on a first surface of the first dielectric layer. The first bonding layer is disposed on the first RDL. The plurality of bottom dies have bonding layers attached to the first bonding layer. The second dielectric layer fills gaps between the bottom dies. The plurality of conductive pillars are disposed in the second dielectric layer without contacting the bottom dies. The second RDL is disposed on the second dielectric layer and is coupled the conductive pillars and a plurality of through silicon vias (TSVs) formed in backsides of the bottom dies. The second bonding layer is disposed on the second RDL. The plurality of top dies have bonding layers attached to the second bonding layer. The third dielectric layer fills gaps between the top dies. The plurality of solder bumps are disposed on a second surface of the first dielectric layer and are coupled to the first RDL through openings of the first dielectric layer.
Another aspect of the present disclosure provides a method for manufacturing a semiconductor package. The method includes receiving a first silicon wafer, forming a first dielectric layer on the first silicon wafer, forming a first RDL on a first surface of the first dielectric layer, forming a first bonding layer on the first RDL, attaching a plurality of bottom dies to the first bonding layer by hybrid bonding, forming a second dielectric layer on the bottom dies, wherein the second dielectric layer fills gaps between the bottom dies, grinding the second dielectric layer and the backsides of the bottom dies so as to reveal a plurality of through silicon vias (TSVs) formed in the backsides of the bottom dies, forming a plurality of conductive pillars in the second dielectric layer without contacting the bottom dies, forming a second RDL on the second dielectric layer, the bottom dies, and the conductive pillars, forming a second bonding layer on the second RDL, attaching a plurality of top dies to the second bonding layer by hybrid bonding, forming a third dielectric layer on the top dies, wherein the third dielectric layer fills gaps between the top dies, grinding the third dielectric layer and backsides of the top dies, attaching a second silicon wafer to the third dielectric layer and the backsides of the top dies, removing the first silicon wafer, and forming a plurality of solder bumps on a second surface of the first dielectric layer.
A more complete understanding of the present disclosure may be derived by referring to the detailed description and claims when considered in connection with the Figures, where like reference numbers refer to similar elements throughout the Figures.
The following description accompanies drawings, which are incorporated in and constitute a part of this specification, and which illustrate embodiments of the disclosure, but the disclosure is not limited to the embodiments. In addition, the following embodiments can be properly integrated to complete another embodiment.
References to “one embodiment,” “an embodiment,” “exemplary embodiment,” “other embodiments,” “another embodiment,” etc. indicate that the embodiment(s) of the disclosure so described may include a particular feature, structure, or characteristic, but not every embodiment necessarily includes the particular feature, structure, or characteristic. Further, repeated use of the phrase “in the embodiment” does not necessarily refer to the same embodiment, although it may.
In order to make the present disclosure completely comprehensible, detailed steps and structures are provided in the following description. Obviously, implementation of the present disclosure does not limit special details known by persons skilled in the art. In addition, known structures and steps are not described in detail, so as not to unnecessarily limit the present disclosure. Preferred embodiments of the present disclosure will be described below in detail. However, in addition to the detailed description, the present disclosure may also be widely implemented in other embodiments. The scope of the present disclosure is not limited to the detailed description, and is defined by the claims.
The first RDL 120 is disposed on a first surface 110A of the first dielectric layer 110. The first bonding layer 130 is disposed on the first RDL 120. Each of the bottom dies 12 has a bonding layer 122, an interconnect layer 124 (including multiple levels of metal lines formed in the back end of line), a device layer 126 (including transistors formed in the front end of line), and a plurality of through silicon vias (TSVs) 128. Generally, a front side of a die is referred to the side with the device layer and the interconnect layer, and a backside of a die is referred to the side with the substrate. In the present embodiment, the bottom dies 12 are attached to the first bonding layer 120 by its bonding layer 122, that is, the bottom dies 12 are disposed on the bonding layer 130 with its backside up. The second dielectric layer 140 fills the gaps between the bottom dies 12. The conductive pillars 150 are disposed in the second dielectric layer 140 without contacting the bottom dies 12.
The second RDL 160 is disposed on the second dielectric layer 140 and coupled to the conductive pillars 150 and TSVs 128 formed in the backsides of the bottom dies 12. The second bonding layer 170 is disposed on the second RDL 160. Each of the top dies 14 has a bonding layer 142, an interconnect layer 144 (including multiple levels of metal lines formed in the back end of line), and a device layer 146 (including transistors formed in the front end of line). The top dies 14 are attached to the second bonding layer 170 by its bonding layer 142. That is, the top dies 14 are disposed on the second RDL 160 and the bonding layer 170 with its backside up. The third dielectric layer 180 fills the gaps between the top dies 14. The solder bumps 190 are disposed on a second surface 110B of the first dielectric layer 110 and coupled to the first RDL 120 through openings of the first dielectric layer 110.
In the present embodiment, the first RDL 120 and the second RDL 160 can provide connecting paths among the bottom dies 12 and the top dies 14 so that each bottom die 12 can be coupled to at least another bottom die 12 and at least one top die 14, and each top die 14 can be coupled to at least another top die 14 and at least one bottom die 12. Furthermore, the top dies 14 and the bottom dies 12 may have front side power delivery network (FSPDN) to receive power form their front sides. For example, the top dies 14 can receive power through the solder bumps 190, the first RDL 120, the conductive pillars 150, and the second RDL 160, and the bottom dies 12 can receive power through the solder bumps 190 and the first RDL 120. However, the present disclosure is not limited thereto. In some embodiments, the bottom dies 12 may have back side power delivery network (BSPDN) to receive power from the solder bumps 190 through the first RDL 120, the conductive pillars 150, the second RDL 160, and the TSVs 128.
In addition, in the present embodiment, the bottom dies 12 and the top dies 14 can be attached on and coupled to the first RDL 120 and the second RDL 160 respectively through the hybrid bonding structures. Since the hybrid bonding technique can provide low resistant connections between the dies and the RDL, resistance of the connecting paths among the bottom dies 12 and the top dies 14 can be significantly reduced. In addition, since no underfill is required for hybrid bonding, the heat generated by the dies 12 and 14 can be dissipated more easily, thereby alleviating the concerns of the thermal issue.
In step S110, a first silicon wafer W1 is received. In some embodiments, the first silicon wafer W1 can be a 300 mm wafer. Afterwards, the first dielectric layer 110 can be formed on the first silicon wafer W1 in step S120 as shown
In step S130, the first RDL 120 is formed on the first surface 110A of the first dielectric layer 110 as shown in
In step S140, the first bonding layer 130 is formed on the first RDL 120 as shown in
During the bonding process between the first bonding layer 130 and the bottom dies 12, the initial bond occurs at the dielectric-to-dielectric interface at room temperature under atmospheric conditions. Subsequently, the copper metal-to-metal connection is formed via annealing and metal diffusion. The hybrid bonding allows smaller pitch than the conventional micro bump bonding. For example, the pitch of the micro bumps may be 10 μm to 20 μm while the pitch of the hybrid bonding may be under 10 μm, such as 9 μm. Furthermore, the hybrid bonding can provide interconnections between the wafer and the dies with lower resistance, so that the efficiency of the signal transmission can be improved.
In step S152, the second dielectric layer 140 can be formed on the bottom dies 12, and the second dielectric layer 140 can fill the gaps between the bottom dies 12 as shown in
In step S156, the conductive pillars 150 can be formed in the second dielectric layer 140 as shown in
In step S160, the second RDL 160 is formed on the second dielectric layer 140, the bottom dies 12, and the conductive pillars 150 as shown in
In step S170, the second bonding layer 170 for receiving the top dies 14 can be formed on the second RDL 160 as shown in
In step S182, the third dielectric layer 180 is formed on the top dies 14 to fill the gaps between the top dies 14 as shown in
In the present embodiments, to avoid copper contamination and ensure the integrity of the semiconductor structure 100 that adopts the hybrid bonding technique, dielectric layers in the second RDL 160 and the third dielectric layer 180 may be made by dielectric material of non-polymer type. Similarly, the first RDL 120 and the second dielectric layer 140 can be made by dielectric material of non-polymer type. For example, the second dielectric layer 140, the third dielectric layer 180, dielectric layers in the first RDL 120, and dielectric layers in the second RDL 160 may include silicon oxide.
In step S190, a second silicon wafer W2 is attached to the third dielectric layer 180 and the backsides of the top dies 14. In some embodiments, the second silicon wafer 194 may be attached to the third dielectric layer 180 and the backsides of the top dies 14 with the aid of bonding dielectric layers. For example, the silicon wafer 194 may have a dielectric layer, and step S190 may be performed with sub-steps for forming another bonding dielectric layer on the third dielectric layer 180 and the backsides of the top dies 14, receiving the second silicon wafer W2, and bonding the bonding dielectric layer on the second silicon wafer W2 and the bonding dielectric layer on the third dielectric layer 180 and the backsides of the top dies 14. As a result, in
In step S192, the first silicon wafer W1 is removed, and the package 100 is flipped upside down as shown in
In some embodiments, a subsequent dicing procedure can be performed to singulate the semiconductor structure 100. In some embodiments, the top dies 14 and the bottom dies 12 can all be core dies that include computation circuits for performing data computations and die-to-die circuits for data transmission between dies.
As shown in
The die-to-die connection circuits 123 and 125, the input/output circuit 127 and the physical interface circuit 129 are placed along four edges of each of the bottom dies 12A1, 12A2, 12B1, and 12B2, and the computation circuit 121 is surrounded by the die-to-die connection circuit 123 and 125, the input/output circuit 127 and the physical interface circuit 129.
Furthermore, in the present embodiments, the two bottom dies 12A1 and 12A2 of the first type and the two bottom dies 12B1 and 12B2 of the second type are disposed in a staggered manner so that every die-to-die connection circuits 123 and 125 of every bottom die is adjacent to another die-to-die connection circuit 123 or 125 of another bottom die. For example, the die-to-die connection circuit 123 of the bottom die 12A1 is adjacent to the die-to-die connection circuit 123 of the bottom die 12B2, and the die-to-die connection circuit 125 of the bottom die 12B2 is adjacent to the die-to-die connection circuit 125 of the bottom die 12A2.
Since the die-to-die connection circuits 123 and 125 of different bottom dies 12A1, 12A2, 12B1, and 12B2 are placed adjacent to each other, the routing paths formed in the first RDL 120 for data sharing between different bottom dies can be shortened, thereby increasing the transmission efficiency and reducing the routing complexity.
Similarly, the top dies 14 of the semiconductor structure 100 may include
In the present embodiment, each of the top dies 14A1, 14A2, 14B1, and 14B2 includes a computation die 141, two die-to-die connection circuits 143 and 145, an input/output circuit 147 and a physical interface circuit 149. The die-to-dic connection circuits 143 and 145, the input/output circuit 147 and the physical interface circuit 149 are placed along four edges of each of the top dies 14A1, 14A2, 14B1, and 14B2, and the computation circuit 141 is surrounded by the die-to-die connection circuit 143 and 145, the input/output circuit 147 and the physical interface circuit 149.
Furthermore, in the present embodiments, the two top dies 14A1 and 14A2 of the first type and the two top dies 14B1 and 14B2 of the second type are disposed in a staggered manner so that every die-to-die connection circuits 143 and 145 of every top die is adjacent to another die-to-die connection circuit 143 or 145 of another top die. For example, the die-to-die connection circuit 143 of the top die 14A1 is adjacent to the die-to-die connection circuit 143 of the bottom die 14B2, and the die-to-die connection circuit 145 of the bottom die 14A1 is adjacent to the die-to-dic connection circuit 145 of the bottom die 14B1.
Since the die-to-die connection circuits 143 and 145 of different top dies 14A1, 14A2, 14B1, and 14B2 are placed adjacent to each other, the routing paths formed in the second RDL 160 for data sharing between different top dies can be shortened, thereby increasing the transmission efficiency and reducing the routing complexity.
In addition, in the present embodiments, from a top view, the die-to-dic connection circuits 143 and 145 of the top dies 14A1, 14A2, 14B1, and 14B2 are aligned with the die-to-die connection circuits 123 and 125 of the bottom dies 12A1, 12A2, 12B1, and 12B2. For example, as shown in
In addition, in some embodiments, to form the first RDL 120 with the conductive traces for connecting the die-to-die connection circuits 123 and 125 of the bottom dies 12A1, 12A2, 12B1, and 12B2, the reticle stitching technique may be adopted. That is, the step S130 may include performing lithography processes by stitching a plurality of reticles (i.e., the lithography masks). Similarly, the first bonding layer 130 may also be formed with mask stitching technique.
For example, currently, the largest reticle available may have a size about 26 mm×33 mm. In such case, if each of the bottom dies 12A1, 12A2, 12B1, and 12B2 has a size about half of the size of the largest reticle, then the first RDL 120 may be formed by stitching two reticles having the largest size. In such case, the semiconductor structure 100 may have a size about 52 mm×33 mm.
Furthermore, in some embodiments, to ensure the traces crossing different reticle regions can be aligned without interrupted, an overlapping region A1 of the reticle region 1201 and an overlapping region A2 of the reticle region 1202 that are overlapping each other may have the same interconnect layout pattern. Also, conductive traces that traverse the boundary between the reticle region 1201 and the reticle region 1202 should traverse the boundary along a direction perpendicular to the boundary. In addition, to ensure the integrity of the first RDL 120, no vias are allowed in the overlapping regions A1 and A2 of the reticle regions 1201 and 1202. In the present embodiment, the same rule can be applied when forming the first bonding layer 130 with stitching masks.
Due to the respective overlapping regions A1 and A2 in the reticle regions 1201 and 1202, the actual sizes of the semiconductor structure 100 may be a bit smaller than the total size of the two reticles. In addition, since the spaces between bottom dies 12A1, 12A2, 12B1, and 12B2 should be preserved for the conductive traces for connecting the adjacent top dies, the size of the bottom dies 12A1, 12A2, 12B1, and 12B2 may be smaller than the half size of the reticle.
In the present embodiments, the second bonding layer 170 and the second RDL 180 may also be formed by stitching masks. For example, the second RDL 180 may include two reticle regions formed by using two different reticles respectively. In addition, the conductive pillars 150 and the solder bumps 190 may also be formed with stitching masks.
Although the RDLs 120 and 160 can be formed by stitching two reticles in the present embodiment, the present disclosure is not limited thereto. In some other embodiments, the designer may stitching even more reticles if necessary.
In the present embodiment, the bottom dies 12A1, 12A2, 12B1, and 12B2 and the top dies 14A1, 14A2, 14B1, and 14B2 may all be chiplets that each includes at least one computation core, so the semiconductor structure 100 can integrate greater computation power of multi-core (e.g., 8 cores) within a small area. In some embodiments, the semiconductor structure 100 can be alone seen as a wafer level semiconductor package, however, in some other embodiments, the semiconductor structure 100 may further be packaged with memory dies.
As shown in
In the present embodiment, the semiconductor structure 100 and the HBMs 220 can be soldered to the interposer 210, and underfills 230 can be applied to protect the soldering structures of the semiconductor structure 100 and the HBMs 220 (e.g., the solder bumps of the semiconductor structure 100 and the HBMs 220). However, the present disclosure is not limited thereto. In some other embodiments, the semiconductor structure 100 and the HBMs 220 may be bonded to the interposer 210 by hybrid bonding so as to further reduce the resistance on the transmission paths and facilitate the thermal dissipation.
Furthermore, a molding compound 240 can also be applied to mold the semiconductor structure 100 and the HBMs 220. Finally, the interposer 210 can be soldered to a substrate 250, and thus, the semiconductor package 200 can be formed in a 2.5D manner. In some embodiments, the interposer 210 may also be formed with the mask stitching technique so as to form the routing paths for connecting the semiconductor structure 100 and the HBMs 200 and the input/output paths for the semiconductor structure 100 and the HBMs 200.
Although the semiconductor structure 100 shown in
As shown in
The input/output dies 34D may include the input/output circuit 127 or 147 shown in
Furthermore, each of the top computing dies 34C may include at least one die-to-die connection circuit 342, and each of the bottom computing dies 32C may include at least one die-to-die connection circuit 322. From a top view, the die-to-die connection circuits 342 of the top computing dies 34C can be aligned with the die-to-die connection circuits 322 of the bottom computing dies 32C. Therefore, the bottom computing dies 32C can be coupled to the corresponding top computing dies 34C vertically with the aids of the second RDL 160 through the die-to-die connection circuits 322 and the die-to-die connection circuits 342. In such case, the semiconductor structure 300 further allows data sharing between the top computing dies 34C and the bottom computing dies 32C.
In the present embodiment, since the input/output dies 34D and the physical interface dies 34E are the top dies that are bonded to the second bonding layer 170 on the second RDL 160, the heat generated by the input/output dies 34D and the physical interface dies 34E can be dissipated through the top of the semiconductor structure 300 more easily, thereby alleviating the thermal concerns of the semiconductor structure 300. Furthermore, by placing the input/output dies 34D and the physical interface dies 34E at the top of the stacking structure also leaves more space for placing the conductive pillars 150 so as to reduce the routing complexity. However, the present disclosure is not limited thereto. In some other embodiments, the input/output dies 34D and the physical interface dies 34E may also be at the bottom of the stacking structure.
In the present embodiment, the top dies 44 may be the computing dies, and each of the top dies 44 may include a computation circuit 441, and a die-to-die connection circuit 443. The computation circuit 441 is for performing the computations required for the applications, and the die-to-die connection circuit 443 is for managing the connection between dies. Each of the bottom dies 42 may include a memory circuit 421, a plurality of die-to-die connection circuits 423, an input/output circuit 425, and a physical interface circuit 427. In some embodiments, the memory circuit 421 may be static random access memory (SRAM).
Furthermore, from a top view, the die-to-die connection circuits 443 of the top dies 44 are aligned with the die-to-die connection circuits 423 of the bottom dies 42. In such case, the top dies 44 can be coupled to the bottom dies 42 through the die-to-die connection circuits 443 and 423 vertically with the aid of the second RDL 160. Therefore, the computation circuit 441 of the top dies 44 can access the memory circuit 421 of the bottom die 42 for more storage spaces, and can utilize the input/output circuit 425 of the bottom die 42 for receiving signals from external and transmitting signals to external. Also, the computation circuit 441 of the top dies 44 may utilize the physical interface circuit 427 of the bottom dies 42 to access the external memory, such as the external dynamic random access memory (DRAM).
In addition, to enable the data sharing between the bottom dies 42, each of the bottom dies 42 may further include a plurality of die-to-die connection circuits 429. The die-to-die connection circuits 429 can be disposed along edges of the bottom dies 42, and by properly placed, each of the die-to-die connection circuits 429 of a bottom die 42 can be adjacent to a die-to-die connection circuits 429 of another bottom die 42. Therefore, the connecting paths between the two adjacent bottom dies 42 can be shortened. In the present embodiment, the input/output circuit 425 and the physical interface circuit 427 can also be disposed along edges of the bottom dies 42. Also, by properly placed, the input/output circuit 425 and the physical interface circuit 427 can be disposed along the outlines of the semiconductor structure 400, thereby facilitating the external routing of the bottom dies 42.
In the present embodiment, each of the semiconductor structures 300 and 400 can be seen as a wafer level package like the semiconductor structure 100. It may also be noted that while the semiconductor structure 100 includes fewer dies of the same function, the semiconductor packages 200 and 300 may include more dies have different functions. In some embodiments, the semiconductor structures 300 and 400 can be seen as a system on wafer (SoW). Also, the semiconductor structures 300 and 400 may have a larger size than the semiconductor structure 100. For example, the size of the semiconductor structures 300 and 400 can be up to 200 mm×200 mm. In such case, each wafer may be used to form only one or two semiconductor structure 300 or 400.
In summary, the semiconductor package and the method for manufacturing the semiconductor package provided by the embodiments of the present disclosure allows the designer to embed more computation power within one package. Also, since the top dies and the bottom dies can be coupled through the RDL in between, flexible routing among different dies can be achieved, thereby improving the efficiency of data sharing and/or data access. Furthermore, since the top dies and the bottom dies can be bonded to the RDL by hybrid bonding, the transmission resistance can be reduced and the thermal issue can be mitigated, which can be a significant improvement for applications requiring massive computations such as artificial intelligence.
Although the present disclosure and its advantages have been described in detail, it should be understood that various changes, substitutions and alterations can be made herein without departing from the spirit and scope of the disclosure as defined by the appended claims. For example, many of the processes discussed above can be implemented in different methodologies and replaced by other processes, or a combination thereof.
Moreover, the scope of the present application is not intended to be limited to the particular embodiments of the process, machine, manufacture, composition of matter, means, methods and steps described in the specification. As one of ordinary skill in the art will readily appreciate from the present disclosure, processes, machines, manufacture, compositions of matter, means, methods or steps, presently existing or later to be developed, that perform substantially the same function or achieve substantially the same result as the corresponding embodiments described herein, may be utilized according to the present disclosure. Accordingly, the appended claims are intended to include within their scope such processes, machines, manufacture, compositions of matter, means, methods and steps.
This application claims the benefit of prior-filed U.S. provisional applications No. 63/387,568, filed on Dec. 15, 2022, and No. 63/580,587, filed on Sep. 5, 2023, which are incorporated by reference in its entirety.
Number | Date | Country | |
---|---|---|---|
63387568 | Dec 2022 | US | |
63580587 | Sep 2023 | US |