Patent Application
20230223402
The present invention is related to electronic circuits, and more particularly, to a three-dimensional (3D) integrated circuit (IC).
Conventional three-dimensional integrated circuit (3D-IC) architectures include a so-called 2.5D architecture and a fully stacked 3D architecture. In a 2.5D architecture, chips are placed side-by-side and interconnected via a horizontal interposer layer. A fully stacked 3D architecture employs chips that are stacked on top of one another. Both architectures use through-silicon vias (TSVs) to connect the metal layers. Compared to 2.5D-IC design, in fully stacked 3D-IC design, chips are connected through bumps and interposers are eliminated. It simplifies the design process and reduces the overall cost.
However, 3D-IC architecture still suffers from some serious drawback. For small changes in IC configurations, the IC layout must be redesigned to provide the feed through channels and allows the signals and power supplies to connect to different chips. For different stack configuration, the IC layouts cannot be reused because the number of the signals and power supplies are varied. The feed through channels form the pyramid shape and occupy significant area in the lower chips, increasing the size and reduces the yield. Moreover, the chips are stacked together, so it is difficult to dissipate heat causing high temperature to significantly degrade overall performance of the IC.
An embodiment provides a 3D (three-dimensional) integrated circuit including a substrate, a first layer on top of the substrate, and a second layer on top of the first layer. The first layer includes a first chip, and a first network bridge formed at a first side of the first chip. The second layer includes a second chip, and a second network bridge formed at a first side of the second chip. The first chip and the first network bridge are coupled to the substrate through bumps. The second chip is coupled to the first chip and the first network bridge through bumps. The second network bridge is coupled to the first network bridge through bumps. The first network bridge and the second network bridge each include a network switch for controlling data transfer and/or power distribution.
Another embodiment provides a 3D integrated circuit including a substrate, a first layer on top of the substrate, and a second layer on top of the first layer. The first layer includes a first chip, and a first network bridge formed at a first side of the first chip, and a second network bridge formed at a second side of the first chip opposite to the first side of the first chip. The second layer includes a second chip, a third network bridge formed at a first side of the second chip, and a fourth network bridge formed at a second side of the second chip opposite to the first side of the second chip. The second chip is coupled to the first chip, the first network bridge and the second network bridge through bumps. The first network bridge and the second network bridge each include a network switch for controlling data transfer and/or power distribution.
Another embodiment provides a 3D integrated circuit including a substrate, a plurality of chips on top of the substrate, and a plurality of network bridges on top of the substrate. A chip of the plurality of chips is on top of a network bridge of the plurality of network bridges, and the chip is coupled to the network bridges through bumps. The chip on top of another chip of the plurality of chips, and the chip is couple to the another chip through bumps. Each of the plurality of network bridges includes a network switch for controlling data transfer and/or power distribution.
These and other objectives of the present invention will no doubt become obvious to those of ordinary skill in the art after reading the following detailed description of the preferred embodiment that is illustrated in the various figures and drawings.
The detailed description set forth below in connection with the appended drawings is intended as a description of various configurations and is not intended to represent the only configurations in which the concepts described herein may be practiced. The detailed description includes specific details for the purpose of providing a thorough understanding of various concepts. However, it will be apparent to those skilled in the art that these concepts may be practiced without these specific details. In some instances, well known structures and components are shown in block diagram form in order to avoid obscuring such concepts.
In a fully stacked 3D architecture, multiple chips are stacked together and connected through TSVs (through silicon-vias). This can improve the overall system performance as well as reduce the cost. For example, fully stacked 3D-ICs are seen as a desirable alternative to overcome interconnect scaling issues that can be a major bottleneck on 2.5D ICs. Fully stacked 3D-ICs, with the advantage of a smaller circuit area, reduce the wire length in each layer. Also, TSVs are implemented for vertical interconnect between chips to reduce long cross-chip wiring.
Instead of using an interposer for data transfer and power distribution as in 2.5D ICs, the 3D-IC stacks chips directly implement data transfer and power distribution in the intermediate chips. Since the thickness of an individual chip is very small, ideally one could mount as many chips as needed. In practice, however, there are several challenges involved in manufacturing 3D-ICs, which restricts the application of 3D-ICs. For small changes in IC configurations, the IC layout must be redesigned to provide the feed through channels and allows the signals and power supplies to connect to different chips.
Please refer to both
The 3D integrated circuit 200 includes a substrate 202, a first layer L1 on top of the substrate 202, and a second layer L2 on top of the first layer L1. The first layer L1 includes chips 204 and 206, and network bridges 208, 210 and 212. The second layer L2 includes chips 214 and 216, and network bridges 218, 220 and 222. The chip 214 is coupled to the chip 204, and the network bridges 208 and 210 through bumps (e.g., solder balls) 226. The chip 216 is coupled to the chip 206, and the network bridges 210 and 212 through bumps 226 with the overlapped region. The overlapped region replaces the interposers in 2.5D-ICs to provide the horizontal connection. Also, the network bridge 218 is coupled to the network bridge 208 through bumps 226; the network bridge 220 is coupled to the network bridge 210 through bumps 226; and the network bridge 222 is coupled to the network bridge 212 through bumps 226. The network bridges 208, 210 and 212 and the chips 204 and 206 are coupled to the substrate 202 through bumps 224 (e.g., solder balls).
In an example, the substrate 202 may be a SiP (System in Package) substrate. The chips 204 and 206 may include logic circuits and the chips 214 and 216 may include memory devices (e.g., registers).
Furthermore, the network bridges 208˜212, 218˜222 may include through-silicon vias (TSVs) 230 to reduce long cross-chip wirings. The network bridges 208˜212, 218˜222 and the chips 204, 206, 214, 216 may include redistribution layers (RDL) on the top surface and/or the bottom surface.
The network bridges 208˜212 and 218˜222 may include network switches 240 (e.g., power control circuit and/or logic switch circuit) for controlling data transfer and/or power distribution. The network switches 240 are coupled to internal wires and can route data and/or power signals in three-dimensions. For example, by controlling the network switches 240, power can be distributed and data can be transferred between the chip 216 and the chip 214 through the network bridge 210. Also, by controlling the network switches 240, power can be distributed and data can be transferred between the chip 204 and the chip 206 through the network bridge 210 and the chips 214, 216. The network switches 240 can be treated as power controllers to support the dynamic voltage scaling and power down mechanism. This architecture without interposers can significantly reduce the length of the power distribution routes to improve heat dissipation. Heat can be transferred more quickly from the center to the edge of the IC package through redistribution layers (RDLs) and through-silicon vias (TSVs). Therefore, this architecture is much simpler than the 3D-IC 100 for thermal management.
In a further example, the network bridges 208 and 210 can act as structural support to allow the larger chip 214 to stack on the smaller chip 204. Moreover, the IC layouts are not required for modification to stack additional layers on the original IC design. Only bumps and RDLs are required to be added for connection of additional layers. This can simplify and accelerate the IC design process.
To explain in more detail, with the network bridges 240, the chip 204 can optionally employ the additional feed through channels for data and/or power distribution. However, the drawback is the larger chip area, high production cost and lower yield. On the other hand, the chip 206 can be fabricated without the TSVs 230. It can rely on the network bridges 240 to support three-dimensional data transfer and/or power distribution. Thus, even without the TSVs 230 in the chip 206, the chip 206 can be integrated in fully stacked 3D-ICs without changes in layout.
The network switches 240 can replace typical crossbar switches in conventional ICs. The data is not limited to two-dimensional (horizontal or vertical) transfer. In fact, the data can be transferred in multiple directions. It can significantly improve the overall data throughput. The same operation can be applied to power distribution. The power can be distributed in multiple directions. Thus, power routing heat dissipation can be made easier.
Moreover, the network switches 240 are not limited only to data transfer and power distribution. It can be implemented to support other functions such as in-memory computing. In the example of in-memory computing, the chip 214 (e.g., memory chip) is stacked on the chip 204 (e.g., logic chip), the network bridges 208 and 210 with the network switches 240 can include cache buffers and act as the memory controller to support the high-speed data transfer. The data from the chip 214 (e.g., memory chip) can be pre-fetched into the network switches 240 for further operations by the chip 204 (e.g., logic chip). The operation results can be temporarily stored in the buffer and then be written back to the chip 214 (e.g., memory chip) later.
In another embodiment, scan chains can be integrated into the network bridges 208˜212 and 218˜222 to test the chips 204, 206, 214, and 216 for KGD (Known Good Die) and examine the chip-to-chip connections to achieve high package yield.
The chip 304 and 306 and the network bridges 308, 310 and 312 are coupled to the substrate 302 through bumps 334 (e.g., solder balls). The chip 314 is coupled to the chip 304 and the network bridge 310 through bumps 336. The network bridge 320 is coupled to the network bridge 310 and the chip 306 through bumps 336. The network bridge 318 is coupled to the network bridge 308 and the chip 304 through bumps 336. The chip 316 is coupled to the network bridge 312 and the chip 306 through bumps 336. The network bridge 328 is coupled to the network bridge 318 through bumps 338. The chip 324 is coupled to the network bridge 318 and the chip 314 through bumps 338. The network bridge 330 is coupled to the network bridge 320 and the chip 314 through bumps 338. The chip 326 is coupled to the network bridge 320 and the chip 316 through bumps 338. The network bridge 332 is coupled to the chip 316 through bumps 338.
In an example, the substrate 302 may be a SiP (System in Package) substrate. The chips 304, 306, 314 and 316 may include logic circuits and the chips 324 and 326 may include memory devices (e.g., registers).
Furthermore, the network bridges 308˜312, 318˜320 and 328˜332 may include through-silicon vias (TSVs) 340 to reduce long cross-chip wirings. The network bridges 308˜312, 318˜320 and 328˜332 and the chips 304, 306, 314, 316, 324 and 326 may include redistribution layers (RDL) on the top surface and/or the bottom surface.
Each of the network bridges 308˜312, 318˜320, and 328˜332 includes a network switch 350 (e.g., power control circuit and/or logic switch circuit) for controlling data transfer and/or power distribution. The operations of the network switches 350 in the 3D-IC 300 are the similar to those in the 3D-IC 200 and are not repeated herein.
Similar to the 3D-IC 200, the network switches 350 can replace typical crossbar switches in conventional ICs. The data is not limited to two-dimensional (horizontal or vertical) transfer. In fact, the data can be transferred in multiple directions. It can significantly improve the overall data throughput. The same operation can be applied to power distribution. The power can be distributed in multiple directions. Thus, power routing heat dissipation can be made easier.
Also similar to the 3D-IC 200, scan chains can be integrated into the network bridges in the 3D-IC 300 to test the chips for KGD (Known Good Die) and examine the chip-to-chip connections to achieve high package yield.
The architecture of the 3D-IC 300 without interposers can significantly reduce the length of the power distribution routes to improve heat dissipation. Heat can be transferred more quickly from the center to the edge of the IC package through redistribution layers (RDLs) and through-silicon vias (TSVs). Therefore, this architecture is much simpler than the 3D-IC 100 for thermal management.
In summary, new 3D-IC architectures are proposed. The network bridges are overlapped with the chips and provide the horizontal connection through the overlapped regions. The data and power supplies can reach different chips through the network bridge design. The new architecture takes advantages of 3D integrated circuit to allow chips with the same function to stack together. The original IC layouts do not need modifications to add additional layers, because the new layers on top can directly couple to the bottom layer through the bumps and redistribution layers (RDL). The network bridges are not limited only to data transfer and power supply. The network bridges can fully utilize the additional silicon area to incorporate other circuits such as memory controllers, cache or memory buffer, scan chains, power switches, and voltage regulators.
Those skilled in the art will readily observe that numerous modifications and alterations of the device and method may be made while retaining the teachings of the invention. Accordingly, the above disclosure should be construed as limited only by the metes and bounds of the appended claims.
