Scalable architecture for reduced cycles across SOC

Abstract

A microelectronic assembly may include a semiconductor wafer having first and second surfaces extending in first and second directions, the semiconductor wafer having network nodes connected to one another via local adjacent connections each extending in only one of the first and second directions, and an interconnection structure comprising a low-loss dielectric material and having first and second opposite surfaces extending in third and fourth directions each oriented at an oblique angle relative to the first and second directions, the interconnection structure having local oblique connections each extending in only one of the third and fourth directions. The semiconductor wafer may be directly bonded to the interconnection structure such that each of the network nodes is connected with at least one of the other network nodes, without use of conductive bonding material, through at least one of the local adjacent connections and at least one of the local oblique connections.

Description

BACKGROUND OF THE INVENTION

The present invention relates to microelectronic packaging and elements thereof and more specifically to an assembly for electrically interconnecting a plurality of processing cores in a common wafer.

The performance of larger processors or high performance computing systems has a diminishing return as the number of processing cores increases. CPU cores and SRAM may compete for limited area on a monolithic die, conventional network grid topologies have a larger number of hops and a higher latency as the number of cores increases, and the resistive-capacitive delay of metal wiring may increase as the number of cores increases, requiring frequent re-timing.

Conventional design architectures limit connections between processor core nodes to orthogonal horizontal and vertical paths, known as Manhattan routes. A hierarchical mesh network topology may improve latency within a network, but such topology is still subject to Manhattan route limitations and increasingly longer routes as the number of cores increases, so such systems are generally limited to connections between neighboring nodes than propagating connections across several nodes. Therefore, despite the advances that have been made in multi-core network grid topologies, there is still a need for further improvements.

BRIEF SUMMARY OF THE INVENTION

One aspect of the disclosure provides a microelectronic assembly that may include a semiconductor wafer having first and second opposite surfaces extending in first and second horizontal orthogonal directions, the semiconductor wafer having a grid of network nodes electrically connected to one another via local adjacent connections each extending in only one of the first and second horizontal orthogonal directions, and an interconnection structure comprising a low-loss dielectric material and having first and second opposite surfaces extending in third and fourth horizontal oblique directions that are each oriented at an oblique angle relative to the first and second directions, the interconnection structure having electrical connections including local oblique connections each extending in only one of the third and fourth horizontal oblique directions. The semiconductor wafer may be directly bonded to the interconnection structure such that each of the network nodes is electrically connected with at least one of the other network nodes. According to some examples, the electrical connection may be formed using direct dielectric bonding, non-adhesive techniques, such as a ZiBond® direct bonding technique, or a DBI® hybrid bonding technique, thermocompression bonding, or other techniques.

Another aspect of the disclosure provides a method including forming local adjacent connections on a surface of a semiconductor wafer, the surface extending in first and second horizontal orthogonal directions, the local adjacent connections electrically connecting a grid of network nodes to one another, the local adjacent connections each extending in only one of the first and second horizontal orthogonal directions. The method may also include forming local oblique connections on a surface of an interconnection structure comprising a low-loss dielectric material, the surface extending in third and fourth horizontal oblique directions that are each oriented at an oblique angle relative to the first and second directions, the local oblique connections each extending in only one of the third and fourth horizontal oblique directions. The method may further include directly bonding the semiconductor wafer to the interconnection structure such that each of the network nodes is electrically connected with at least one of the other network nodes, for example, using a direct bonding technique.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a diagrammatic plan view of the connections between adjacent nodes in a conventional multi-core processor die.

FIG. 2 is a diagrammatic plan view of the connections between adjacent nodes in a multi-core processor die according to one embodiment of the invention.

FIG. 3 is a diagrammatic plan view of a communication node design of a 64-core processor die.

FIG. 4 is a diagrammatic plan view of connections between nearby nodes in conventional design architecture.

FIG. 5 is a diagrammatic plan view of connections between nearby nodes in a design architecture according to an embodiment of the invention.

FIG. 5A is a diagrammatic plan view of horizontal connections between nearby nodes in the design architecture of FIG. 5.

FIG. 5B is a diagrammatic plan view of diagonal connections between nearby nodes in the design architecture of FIG. 5.

FIG. 6 is a diagrammatic plan view of a plurality of reticles overlying a semiconductor wafer, the reticles configured to create an interconnection pattern configured in a Manhattan routing.

FIG. 7 is a diagrammatic plan view of a plurality of reticles overlying a low-loss interposer, the reticles configured to create an interconnection pattern at an oblique angle to a Manhattan routing.

FIG. 8 is a diagrammatic plan view of a portion of the low-loss interposer of FIG. 7 after the oblique interconnection pattern is created.

FIG. 9 is a diagrammatic plan view of a portion of the low-loss interposer of FIG. 7 rotated to a position that aligns with the corresponding semiconductor wafer of FIG. 6.

FIG. 10 is a diagrammatic plan view of a portion of a low-loss interposer having a non-45 degree oblique interconnection pattern.

FIGS. 11-11A are a side sectional views of example packages including a semiconductor having a Manhattan routing and a low-loss interposer having an interconnection at an oblique angle to the Manhattan routing.

FIG. 12 illustrates an example system according to aspects of the disclosure.

DETAILED DESCRIPTION

As used in this disclosure with reference to an element having a planar surface, a statement that an electrically conductive element is “at” a surface of a substrate indicates that, when the substrate is not assembled with any other element, the electrically conductive element is available for contact with a theoretical point moving in a direction perpendicular to the surface of the substrate toward the surface of the substrate from outside the substrate. Thus, a terminal or other conductive element which is at a surface of a substrate may project from such surface; may be flush with such surface; or may be recessed relative to such surface in a hole or depression in the substrate. In some embodiments, the conductive element may be attached to the surface or may be disposed in one or more layers of dielectric coating on the said surface.

In this disclosure, first and second orthogonal directions X and Y are referred to herein as “horizontal” or “lateral” directions, whereas the directions perpendicular to a plane defined by the X and Y directions, such as a third direction Z, are referred to herein as upward or downward directions and are also referred to herein as the “vertical” directions. The directions referred to herein are in the frame of reference of the structures referred to. Thus, these directions may lie at any orientation to the normal or gravitational frame of reference. A statement that one feature is disposed at a greater height “above a surface” than another feature means that the one feature is at a greater distance in the same orthogonal direction away from the surface than the other feature. Conversely, a statement that one feature is disposed at a lesser height “above a surface” than another feature means that the one feature is at a smaller distance in the same orthogonal direction away from the surface than the other feature.

As illustrated in FIG. 1, a conventional multi-core processor wafer 10 may have a plurality of adjacent nodes 14a, 14b, 14c, and 14d (collectively nodes 14) arranged at the vertices of a square. Each of the nodes 14 may be electrically connected with adjacent ones of the nodes via an adjacent connection 15, and each of the nodes may be electrically connected with an opposite one of the nodes via an opposite connection 16. In a conventional node arrangement having an orthogonal or Manhattan routing, each of the adjacent connections 15 extends in one of the horizontal orthogonal directions X or Y, and each of the opposite connections 16 only has portions extending in the directions X and Y. In one example in which the adjacent nodes 14 are horizontally and vertically spaced apart from one another by 8 mm, each of the adjacent connections 15 will have a length of 8 mm, and each of the opposite connections 16 will have a length of 16 mm Limitations in lithography technology currently limit routing in CMOS wafers to electrical connections extending in orthogonal X and Y directions, as shown in FIG. 1.

In contrast, as illustrated in FIG. 2, a multi-core processor wafer 20 according to one embodiment of the invention may have a plurality of adjacent nodes 24a, 24b, 24c, and 24d (collectively nodes 24) arranged at the vertices of a square. Each of the nodes 24 may be electrically connected with adjacent ones of the nodes via an adjacent connection 25, and each of the nodes may be electrically connected with an opposite one of the nodes via an oblique connection 26. In a node arrangement according to the invention, each of the adjacent connections 25 extends in one of the orthogonal horizontal directions X or Y, and each of the oblique connections 26 extends in oblique horizontal directions V or W. As shown, the oblique directions V and W are rotated 45 degrees from the orthogonal directions X and Y, although in other embodiments (e.g., FIG. 10), other oblique rotational orientations may be used. In one example in which the adjacent nodes 24 are horizontally and vertically spaced apart from one another by 8 mm, each of the adjacent connections 25 will have a length of 8 mm, and each of the oblique connections 26 will have a length of 11.3 mm, which is 29.3% less than the 16 mm length of the opposite connections 14 of FIG. 1.

As illustrated in FIG. 3, a conventional 64-core processor wafer arrangement is shown on a wafer 30 that has sixteen nodes 34, each node arranged at an intersection of four of the cores 31. The wafer 30 may be a semiconductor wafer that is implemented in one of various semiconductor materials such as silicon, silicon-germanium, indium phosphide, and gallium arsenide or one or more other Group III-V semiconductor compounds or Group II-VI semiconductor compounds, etc. The wafer 30 may have lengths in the orthogonal X and Y directions of 32 mm, for example. The wafer 30 may have layers of electrical connections configured to function in a hierarchical mesh network topology, including a core layer in which each node 34 is connected to each of the four cores 31 that are in contact with the node, and a network layer vertically adjacent to the core layer (in a Z direction orthogonal to the X and Y directions) a that contains connections between the nodes. For example, table 1 below shows the relationship between nodes 34 and cores 31 in a conventional hierarchical switch architecture, in which one node is required for every four cores, one hop is required for every log₄(nodes), and three switch ports are required for every hop.

	TABLE 1
	CPU	Switch		Switch
	Cores	Nodes	Hops	Ports
	(c)	(n)	(h)	(p)
	16	4	1	3
	64	16	2	6
	256	64	3	9
	1024	256	4	12

FIG. 4 illustrates a conventional 64-core processor wafer arrangement that is shown on a wafer 40 that has sixteen nodes 44, each node arranged at an intersection of four cores, in the manner shown in FIG. 3. The wafer 40 may have layers of electrical connections, including a network layer 42 that can be used to connect the nodes 44 to one another. Similar to the architecture shown in FIG. 1, the network layer 42 may have local adjacent connections 45 between adjacent ones of the nodes 44 and local opposite connections 46 between opposite ones of the nodes. However, the network layer 42 may also have longer unidirectional connections 47 that extend between two non-adjacent nodes 44 in one of the orthogonal horizontal directions X or Y and longer bidirectional connections 48 that extend between two non-adjacent modes only having portions extending in the directions X and Y. In one example in which the adjacent nodes 44 are horizontally and vertically spaced apart from one another by 8 mm, each of the local adjacent connections 45 will have a length of 8 mm, each of the local opposite connections 46 will have a length of 16 mm, each of the longer unidirectional connections 47 will have a length of 16 mm, and each of the longer bidirectional connections 48 will have a length of 32 mm.

FIGS. 5, 5A, and 5B illustrate a microelectronic assembly 50 including a wafer 60 and an interconnection structure 70. The wafer 60 includes a 64-core processor wafer arrangement according to an embodiment of the invention. The wafer 60 has sixteen nodes 64, each node arranged at an intersection of four cores, in the manner shown in FIG. 3. The wafer 60 may have unitary crystalline structure of semiconductor material, or it may be a reconstituted wafer. The wafer 60 may have an orthogonal network layer 62 of electrical connections (shown separately in FIG. 5A) that can connect the nodes 64 to one another along directions of a first orthogonal set of directions X, Y. The interconnection structure 70 may have an oblique network layer 72 (shown separately in FIG. 5B) that can connect the nodes 64 to one another along directions of a second set of directions V, W that are each oriented at an oblique angle to the first orthogonal set of directions.

Referring to FIGS. 5 and 5A, the orthogonal network layer 62 may have local adjacent connections 65 between adjacent ones of the nodes 64 and longer unidirectional connections 67 that extend between two non-adjacent nodes in one of the orthogonal horizontal directions X or Y. The orthogonal network layer 62 may be contained within the wafer 60. In one example in which the adjacent nodes are horizontally and vertically spaced apart from one another by 8 mm, each of the local adjacent connections 65 will have a length of 8 mm, and each of the longer unidirectional connections 67 will have a length of 16 mm.

Referring to FIGS. 5 and 5B, similar to the architecture shown in FIG. 2, the oblique network layer 72 may have local oblique connections 76 between opposite ones of the nodes. Each of the local oblique connections 76 extends in oblique horizontal directions V or W that are rotated 45 degrees from the orthogonal directions X and Y, although in other embodiments (e.g., FIG. 10), other oblique rotational orientations may be used. The oblique network layer 72 may also have longer oblique connections 78 that extend between two non-adjacent nodes 64 in the oblique horizontal directions V or W.

In an example in which the adjacent nodes 64 are horizontally and vertically spaced apart from one another by 8 mm, each of the local oblique connections 76 will have a length of 11.3 mm, and each of the longer oblique connections 78 will have a length of 22.6 mm Therefore, the local oblique connections 76 of 11.3 mm are each 29.3% less than the 16 mm length of the local opposite connections 46 of FIG. 4, and the longer oblique connections 78 of 22.6 mm are each 29.3% less than the 32 mm length of the longer bidirectional connections 48 of FIG. 4.

The oblique network layer 72 may extend within the interconnection structure 70 that is affixed to an active surface of the wafer 60. The interconnection structure 70 may comprise a low-loss directly-bonded interconnect layer or layers that permit the oblique routing of the local oblique connections 76 and the longer oblique connections 78. Suitable low-loss materials for the oblique network layer 72 may include glass, quartz, sapphire, or ZIF. The use of low-loss materials for the interconnection structure 70 that includes the oblique network layer 72 permits longer-distance connections to experience shorter latency than the same connections in semiconductor material.

The interconnection structure 70 may be bonded in a stacked arrangement with the wafer 60 using various bonding techniques, including using direct dielectric bonding, non-adhesive techniques, such as a ZiBond® direct bonding technique, or a DBI® hybrid bonding technique, both available from Invensas Bonding Technologies, Inc. (formerly Ziptronix, Inc.), a subsidiary of Xperi Corp. (see for example, U.S. Pat. Nos. 6,864,585 and 7,485,968, which are incorporated herein in their entirety). According to some examples, thermocompression bonding may be used. This process may enable the interconnection structure 70 to bond to the wafer 60 with adjacent electrical connections at an extremely fine pitch. For example, the connection may be at a pitch as low as approximately 1μ-10μ. In such a direct bonding example, a surface of the interconnection structure 70 may be laminated onto a confronting exposed front surface of the wafer 60, and heat and pressure may be used to bond the confronting surfaces to one another.

Referring to FIG. 6, the orthogonal network layer 62 may be formed in the wafer 60 using reticles 52 as an aligner mask. An arrangement of the reticles 52 may be placed overlying the wafer 60 in a pattern extending in the first orthogonal set of directions X, Y. As shown in FIG. 6, each reticle 52 is rectangular in shape, but in other examples, each reticle may have a square shape. Lithography is then performed on the wafer 60 to form the local adjacent connections 65 and the longer unidirectional connections 67 of the orthogonal network layer 62, the connections extending in the first orthogonal set of directions X, Y relative to the lithography equipment.

Referring now to FIG. 7, the oblique network layer 72 may be formed in the interconnection structure 70 using the reticles 52 as an aligner mask. An arrangement of the reticles 52 may be placed overlying the interconnection structure 70 in a pattern extending in the first orthogonal set of directions X, Y. As shown in FIG. 7, each reticle 52 is rectangular in shape, but in other examples, each reticle may have a square shape. Before the lithography is performed, the interconnection structure 70 with the reticles 52 thereon may be rotated relative to the lithography equipment, by an oblique angle such as 45 degrees. Then, lithography may performed to form connections extending in the second set of directions V, W that are obliquely oriented relative to the X, Y orientation of the reticles.

The resulting connection patterns in the interconnection structure 70 relative to one of the reticles 52 are shown in FIG. 8, in which the local oblique connections 76 and the longer oblique connections 78 of the oblique network layer 72 extend in the second set of horizontal directions V, W, while the example reticle 52 extends in the first set of horizontal directions X, Y.

Accordingly, referring to FIG. 9, after the local oblique connections 76 and the longer oblique connections 78 are formed in the interconnection structure 70, the interconnection structure may be rotated relative to the wafer 60, so that the areas that were previously covered by the reticles 52 are in alignment between the wafer and the interconnection structure. The combined pattern of connections will be as shown in FIG. 5, in which the local adjacent connections 65 and the longer unidirectional connections 67 of the orthogonal network layer 62 extend in the first orthogonal set of directions X, Y, and in which the local oblique connections 76 and the longer oblique connections 78 of the oblique network layer 72 extend in the second oblique set of directions V, W.

FIG. 10 illustrates connections in a microelectronic assembly 50′ that is a variation of the microelectronic assembly 50, but having a non-45 degree oblique interconnection pattern between the local adjacent connections 65′ and the local oblique connections 76′. For example, rather than having the second oblique set of directions V, W be orthogonal relative to one another, the directions V′ and W′ can each be obliquely oriented at the same angle relative to the horizontal direction X, which need not be 45 degrees. As shown in FIG. 10, each of the directions V′ and W′ may be angled at 30 degrees relative to the horizontal direction X, such that the directions V′ and W′ are offset from each other by 120 degrees. The example shown in FIG. 10 is only one possibility. The directions V′ and W′ may be angled at other oblique angles relative to the horizontal direction X, such as 15°, 20°, 30°, 50°, 60°, 75°, or other angles not listed here.

FIG. 11 illustrates one example of a microelectronic package 80 including the wafer 60 and the interconnection structure 70 mounted to a substrate 81. The microelectronic package 80 may have a core layer 90 that is configured to function in a hierarchical mesh network topology with the wafer 60 (that serves as the network layer). An interposer 91 may be disposed adjacent to the core layer 90, and the interconnection structure 70 may be disposed between the interposer 91 and the wafer 60.

The core layer 90, the interposer 91, and the wafer 60 may each have electrical connections extending in the first set of orthogonal horizontal directions X and Y, and the interconnection structure 70 may have electrical interconnections extending in the second set of oblique horizontal directions V and W, similar to the connections described above with regard to the microelectronic assembly 50. Expansion wafers 92 similar to the base wafer 60 may be stacked above the base wafer 60. In the embodiment shown, the core layer 90, the base wafer 60, and the expansion wafers 92 may together function as a hierarchical mesh network.

The core layer 90 may be flip-chip mounted to a first surface 82 of the substrate 81 via conductive bumps 84, which may be spaced from one another at a 150μ pitch, for example. A second surface 83 of the substrate 81 may have conductive terminals 85 exposed thereat that are configured for electrical connection with a component external to the microelectronic package 80. Each of the core layer 90, the interposer 91, the interconnection structure 70, the base wafer 60, and the expansion wafers 92 may be directly bonded to one another in a vertical stack overlying the substrate 80, using any of the various bonding techniques described above with reference to the microelectronic assembly 50. Through-silicon vias, or TSVs (not shown), may be used for inter-wafer electrical interconnection among the aforementioned layers, wafers, and structures. FIG. 11A illustrates an embodiment where the core layer 90, the interconnection structure 70, the base wafer 60, and the expansion wafers 92 may be directly bonded to one another in a vertical stack overlying the substrate 80, without an interposer.

The microelectronic assemblies and microelectronic packages described above with reference to FIGS. 1-11 above can be utilized in construction of diverse electronic systems, such as the system 100 shown in FIG. 12. For example, the system 100 in accordance with a further embodiment of the invention includes a plurality of modules or components 106 such as the microelectronic assemblies 50 and 50′ and microelectronic package 80 described above, in conjunction with other electronic components 108, 110 and 111.

In the exemplary system 100 shown, the system can include a circuit panel, motherboard, or riser panel 102 such as a flexible printed circuit board, and the circuit panel can include numerous conductors 104, of which only one is depicted in FIG. 12, interconnecting the modules or components 106, 108, 110 with one another. Such a circuit panel 102 can transport signals to and from each of the microelectronic packages and/or microelectronic assemblies included in the system 100. However, this is merely exemplary; any suitable structure for making electrical connections between the modules or components 106 can be used.

In a particular embodiment, the system 100 can also include a processor such as the semiconductor chip 108, such that each module or component 106 can be configured to transfer a number N of data bits in parallel in a clock cycle, and the processor can be configured to transfer a number M of data bits in parallel in a clock cycle, M being greater than or equal to N.

In the example depicted in FIG. 12, the component 108 is a semiconductor chip and component 110 is a display screen, but any other components can be used in the system 100. Of course, although only two additional components 108 and 111 are depicted in FIG. 12 for clarity of illustration, the system 100 can include any number of such components.

Modules or components 106 and components 108 and 111 can be mounted in a common housing 101, schematically depicted in broken lines, and can be electrically interconnected with one another as necessary to form the desired circuit. The housing 101 is depicted as a portable housing of the type usable, for example, in a cellular telephone or personal digital assistant, and screen 110 can be exposed at the surface of the housing. In embodiments where a structure 106 includes a light-sensitive element such as an imaging chip, a lens 111 or other optical device also can be provided for routing light to the structure. Again, the simplified system shown in FIG. 12 is merely exemplary; other systems, including systems commonly regarded as fixed structures, such as desktop computers, routers and the like can be made using the structures discussed above.

It will be appreciated that the various dependent claims and the features set forth therein can be combined in different ways than presented in the initial claims. It will also be appreciated that the features described in connection with individual embodiments may be shared with others of the described embodiments. For example, the microelectronic package 80 shown in FIG. 11 may be modified to omit the expansion layers 92. Also, although the wafers herein are described as semiconductor wafer, in other embodiments, the wafers herein such as the wafer 60 may be a semiconductor die that is directly bonded to a semiconductor wafer and/or an interconnection structure such as the interconnection structure 70.

Although the invention herein has been described with reference to particular embodiments, it is to be understood that these embodiments are merely illustrative of the principles and applications of the present invention. It is therefore to be understood that numerous modifications may be made to the illustrative embodiments and that other arrangements may be devised without departing from the spirit and scope of the present invention as defined by the appended claims.

Claims

1. A microelectronic assembly, comprising: a semiconductor wafer having first and second opposite surfaces, the semiconductor wafer having a grid of sixteen network nodes electrically connected to one another via local adjacent connections each extending in only one of first and second horizontal orthogonal directions, the sixteen network nodes connecting to a plurality of processor cores of a hierarchical mesh network; an interconnection structure comprising a low-loss dielectric material and having first and second opposite surfaces, the interconnection structure having electrical connections including local oblique connections each extending in only one of third and fourth horizontal oblique directions that are each oriented at an oblique angle relative to the first and second directions; andwherein the first surface of the semiconductor wafer is directly hybrid bonded, without using an adhesive, to the first surface of the interconnection structure such that each of the network nodes is electrically connected with at least one of the other network nodes through at least one of the local adjacent connections and at least one of the local oblique connections,wherein, the hierarchical mesh network is configured such that a maximum of two hops is needed for an electrical signal to travel from any one of the network nodes, through the interconnection structure, to any other one of the network nodes, andwherein the maximum of two hops is a maximum of two direct electrical signal connections extending between the any one of the network nodes to the any other one of the network nodes.

2. The microelectronic assembly of claim 1, wherein the semiconductor wafer further includes longer unidirectional connections each extending in only one of the first or second horizontal orthogonal directions, each of the longer unidirectional connections extending between non-adjacent ones of the network nodes that are spaced apart from one another along only one of the first or second horizontal orthogonal directions.

3. The microelectronic assembly of claim 2, wherein the interconnection structure further includes longer oblique connections each extending in only one of the third or fourth horizontal oblique directions, each of the longer oblique connections providing at least a portion of a respective electrical connection between non-adjacent ones of the network nodes that are spaced apart from one another along only one of the third or fourth horizontal oblique directions.

4. The microelectronic assembly of claim 1, wherein the oblique angle is 45 degrees.

5. The microelectronic assembly of claim 1, wherein the low-loss dielectric material is one or more of glass, quartz, sapphire, benzocyclobutene (BCB) or ZIF.

6. The microelectronic assembly of claim 1, further comprising a core semiconductor layer having the plurality of processor cores therein, the core semiconductor layer being electrically connected with the semiconductor wafer and the interconnection structure, wherein the core semiconductor layer, the semiconductor wafer, and the interconnection structure together are configured to function as the hierarchical mesh network.

7. The microelectronic assembly of claim 6, wherein the first and second surfaces of the interconnection structure are directly bonded to the semiconductor wafer and the core semiconductor layer, respectively.

8. The microelectronic assembly of claim 1, wherein the low-loss dielectric material is incompatible with processing used for semiconductor elements having active semiconductor devices therein.

9. A microelectronic package including the microelectronic assembly of claim 7, further comprising a package substrate electrically connected with the core semiconductor layer, the semiconductor wafer, and the interconnection structure, the package substrate having a plurality of terminals at a surface thereof, the terminals being configured for electrical connection with a component external to the microelectronic package.

10. A method of forming a microelectronic assembly, comprising: forming local adjacent connections on a surface of a semiconductor wafer, the surface extending in first and second horizontal orthogonal directions, the local adjacent connections electrically connecting a grid of sixteen network nodes to one another, the local adjacent connections each extending in only one of the first and second horizontal orthogonal directions;forming local oblique connections on a surface of an interconnection structure comprising a low-loss dielectric material, the surface extending in third and fourth horizontal oblique directions that are each oriented at an oblique angle relative to the first and second directions, the local oblique connections each extending in only one of the third and fourth horizontal oblique directions;directly hybrid bonding, using a non-adhesive technique, a first surface of the semiconductor wafer to a first surface of the interconnection structure, such that the first surface of the semiconductor wafer is in surface-to-surface contact with the first surface of the interconnection structure, and such that each of the network nodes is electrically connected with at least one of the other network nodes through at least one of the local adjacent connections and at least one of the local oblique connections; andinterconnecting, through a combination of the semiconductor wafer and the interconnection structure, a plurality of processor cores, wherein the plurality of processor cores, the semiconductor wafer, and the interconnection structure are together configured to function as a hierarchical mesh network,wherein, the hierarchical mesh network is configured such that a maximum of two hops is needed for an electrical signal to travel from any one of the network nodes, through the interconnection structure, to any other one of the network nodes, andwherein the maximum of two hops is a maximum of two direct electrical signal connections extending between the any one of the network nodes to the any other one of the network nodes.

11. The method of claim 10, wherein the low-loss dielectric material is incompatible with processing used for semiconductor elements having active semiconductor devices therein.

12. The method of claim 11, further comprising, before the directly bonding, rotationally aligning the surface of the interconnection structure relative to the surface of the semiconductor wafer such that the third and fourth horizontal oblique directions are each oriented at the oblique angle relative to the first and second horizontal orthogonal directions.

13. The method of claim 11, wherein the forming of the local oblique connections on the surface of an interconnection structure is performed by placing a reticle overlying a portion of the surface of the interconnection structure, and using lithography to remove linearly-extending portions of a mask overlying the surface of the interconnection structure, the linearly-extending portions extending at oblique angles relative to outer peripheral boundaries of the reticle.

14. The method of claim 13, wherein the reticle has a rectangular shape, and the semiconductor wafer has a maximum length and a maximum width in the respective first and second horizontal orthogonal directions, the maximum length and the maximum width being less than or equal to than a length and width of a maximum square overlap area resulting from overlying the reticle with an identical reticle rotated to the oblique angle relative to the reticle.

15. The method of claim 11, further comprising forming longer unidirectional connections on the surface of the semiconductor wafer, the longer unidirectional connections each extending in only one of the first or second horizontal orthogonal directions, each of the longer unidirectional connections extending between non-adjacent ones of the network nodes that are spaced apart from one another along only one of the first or second horizontal orthogonal directions.

16. The method of claim 15, further comprising forming longer oblique connections on the surface of the interconnection structure, the longer oblique connections each extending in only one of the third or fourth horizontal oblique directions, each of the longer oblique connections providing at least a portion of a respective electrical connection between non-adjacent ones of the network nodes that are spaced apart from one another along only one of the third or fourth horizontal oblique directions.

17. The method of claim 11, further comprising directly bonding and electrically connecting a plurality of semiconductor expansion layers to the semiconductor wafer, without the use of the conductive bonding material.

18. The method of claim 11, wherein the plurality of processor cores are formed in a core semiconductor layer, further comprising joining the core semiconductor layer to the semiconductor wafer or the interconnection structure.