Apparatus with multiple optical duplexing and parallel connector adapters, multi-port passive photonic circuit chip and structured cable system

Abstract

The utility model relates to a device with a plurality of optical duplex and parallel connector adapters, a multi-port passive photonic circuit chip and a structured cable system. An apparatus having multiple optical duplexing and parallel connector adapters, such as MPO connectors and LC adapters, some of which are connected to network devices in a network while other adapters are connected to servers or processing units, such as GPUs, combining internal photonic circuits with a grid. The optical path of each transmitter and receiver is matched so as to provide an appropriate optical connection from the transmitting fiber to the receiving fiber, where any complex network topology can be achieved.

Description

Technical Field

[0001] Disclosed is a device and method for improving the scalability of a data center network using a mesh network topology, switches with various cardinalities, hierarchies, and oversubscription ratios. The device and method disclosed in the utility model reduces the number of manual network connections, simplifies wiring installation, and improves the flexibility and reliability of a data center at a reduced cost. Background Art

[0002] The importance of using optical fiber to transmit communication signals has grown rapidly due to its high bandwidth, low attenuation and other significant advantages, including radiation resistance, small size and light weight. Data center architectures using optical fiber are evolving to meet global traffic demands and the increasing number of users and applications. The rise of cloud data centers, especially hyperscale clouds, has significantly changed enterprise information technology (IT) business structures, network systems and topologies. In addition, the requirements of cloud data centers are affecting technology roadmaps and standardization.

[0003] The widespread adoption of server virtualization and advances in data processing and storage technologies have led to the growth of east-west traffic within data centers. The traditional three-layer switch architecture, including the core layer, aggregation layer, and access layer (CAA), cannot provide the low and balanced latency channels required for east-west traffic. In addition, since the CAA architecture utilizes the spanning tree protocol to disable redundant paths and build a loop-free topology, the CAA architecture does not fully utilize network capacity.

[0004] The Folded Clos network (FCN) or Spine-and-Leaf architecture is a topology that is better suited to overcome the limitations of three-tier CAA networks. The Clos network is a multi-stage circuit-switched network proposed by Charles Clos in 1953 that has become very important today due to the use of a complex optical interconnect topology. The Folded Clos network topology utilizes two types of switching nodes: Spine and Leaf. Each Spine is connected to each Leaf. The network can be scaled horizontally to enable communication between a large number of servers while minimizing latency and non-uniformity by simply adding more Spine and Leaf switches.

[0005] Figure 1 shows an example of two FCNs with a similar number of hosts using different radixes and levels. The higher radix (32 in this example) connects 32 leaf switches in a two-tier network, as shown in part (A) of the figure. A two-tier FCN provides the lowest latency, but at the expense of a denser network (512 interconnects). By using a three-tier network, the interconnect layout can become less complex or more structured. However, this requires more switches and introduces more latency into the network. Over the past few years, the need for flatter networks to handle the growing amount of traffic between machines has favored an increase in the radix of switch application-specific integrated circuits (ASICs). Currently, switch ASIC radixes can handle 256 ports, each with a speed of 100Gb / s. Higher radix ASICs are expected in the future.

[0006] Based on the industry telecommunications infrastructure standard TIA 942-A, leaf switches and spine switches can be located tens or hundreds of meters apart. Typically, spine switches are located in the main distribution area (MDA), while leaf switches are located in the equipment distribution area (EDA) or horizontal distribution area (HDA).

[0007] The architecture has been proven to deliver high bandwidth and low latency (only two hops to reach the destination) with low oversubscribed connectivity. However, for a large number of switches, the spine-leaf architecture requires a complex mesh with a large number of fibers and connectors, which increases the cost and complexity of the installation.

[0008] Future data centers will require more flexible and adaptable networks than the traditional grids implemented today to accommodate highly distributed computing, machine learning (ML) training loads, advanced virtualization, and data replication.

[0009] Traditionally, mesh switching networks such as those shown in FIG. 1 are implemented in patch panels that use hundreds or thousands of jumper connections to deploy network topologies. More recently, the use of transpose boxes such as those shown in the prior art and listed RS can help deploy these networks while reducing installation errors. Transpose boxes use multiple duplex fiber connectors or optical flex circuits to implement a portion of a network mesh within the box. Optical flex circuits are arrays of wiring optical fibers embedded on two layers of polymer that can provide a versatile mesh deployment on a flexible substrate.

[0010] Better ways to implement meshes using planar lightwave circuits (PLCs) have been considered to achieve today's complex mesh topologies. Currently, most PLC approaches involve lithography and etching on silicon wafers. However, waveguides fabricated using lithography limit device fabrication to 2D configurations. Therefore, using lithographic processes, implementing a mesh similar to the topology shown in Figure 1 would require multiple crossovers, resulting in high losses and crosstalk.

[0011] A newer fabrication method is direct laser writing of waveguides on glass (DLWW) using a femtosecond laser.

[0012] Femtosecond lasers produce high-intensity pulses lasting 10 or 100 femtoseconds, which can effectively nonlinearly absorb a few near-infrared (NIR) photons, thereby changing the properties of the glass. Nonlinear absorption of NIR pulses is less destructive than UV absorption and facilitates the use of versatile optical devices to control the spatial distribution of intensity in three dimensions within the glass. During the past decade, DLWW has made significant progress in reducing waveguide distortion and other defects that reduce propagation and coupling losses [4].

[0013] These latest technologies for DLWW enable complex 3D photonic circuits (PCs) and facilitate rapid maskless prototyping. DLWW can fabricate waveguides with similar properties to single-mode optical fibers (SMFs), which is beneficial for reducing coupling losses. DLWW supports 3D geometries that virtually eliminate waveguide crossings and crosstalk between channels.

[0014] DLWW can be a useful technique for fabricating 3D multi-channel optical meshes. However, DLWW has two major limitations that require careful design of the waveguides: propagation losses that are higher or at best similar to those arising from the photolithographic process mentioned earlier and bending losses induced by the relatively low refractive index contrast of the waveguides.

[0015] A mesh DLWW small form factor with low loss and negligible crosstalk is disclosed here to overcome the above limitations. Utility Model Content

[0016] An apparatus with multiple optical duplex and parallel connector adapters (such as MPO connectors and LC adapters), some of which connect to network devices in the network and others to servers or processing units (such as GPUs) combines internal photonic circuits with the mesh. The optical paths of each transmitter and receiver are matched to provide the appropriate optical connection from the transmitting fiber to the receiving fiber, where complex arbitrary network topologies can be implemented. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] FIG1(A) shows a two-stage FCN base 32.

[0018] FIG1(B) shows a three-level FCN cardinality 8.

[0019] FIG. 2 shows a top view of a submodule 500 showing an interconnection arrangement. picture.

[0020] FIG. 3 shows twelve of the twenty-four alternative grids that may be implemented in submodule 500 .

[0021] FIG. 4( a ) shows a top view of the DLWW optical chip 700 picture.

[0022] FIG. 4( b ) shows a side view of the DLWW optical chip 700 , showing the different layers into which the waveguides are written to avoid crossing.

[0023] FIG. 5 shows a top view of the DLWW optical chip 700. picture.

[0024] FIG. 6 illustrates an exemplary module 900 showing the connection of the DLWW embodiment 700 to two fiber arrays and to the multi-fiber ports of the module.

[0025] FIG. 7 illustrates an exemplary module 400 showing the connection of a DLWW embodiment 700 to a fiber array and to a multi-fiber port of the module.

[0026] FIG. 8 illustrates a simple method of using module 400 to implement a network with 16 leaf switches and up to 16 spine switches. DETAILED DESCRIPTION

[0027] DLWW is a relatively new technology that enables the fabrication of optical grid waveguides with refractive index contrast and a circular shape closer to single-mode fiber (SMF), making the waveguide dimensions comparable, thereby reducing coupling losses. In addition, the 3D geometry achieved by DLWW virtually eliminates waveguide crossings and thus crosstalk between channels. However, the technology has limitations such as propagation and bending losses, tight tolerances, and the need for careful waveguide design.

[0028] An advantage of the invention disclosed in this document is that it provides a small form factor grid capable of supporting communication signals transmitted over multiple channels with negligible crosstalk or noise losses and limited additional losses beyond the designed split ratio.

[0029] The present invention focuses on transmitters with four and eight channels, such as 400GBASE-DR4. Several embodiments will be shown in this section. Please note that all drawings are not drawn to scale in order to emphasize the main features of the device.

[0030] FIG. 2 shows a front view of a disclosed embodiment 500, a photonic circuit chip, which is a key element to facilitate the deployment, reshaping and expansion of optical networks. This embodiment is used to generate an interconnected grid of four groups of waveguides. As shown in the figure, eight parallel ports 510, 515, 520, 525, 550, 555, 560 and 565 are distributed on the front and rear sides of this embodiment. Each port consists of eight waveguides labeled from a to h. The grid of submodule 500 can be implemented in twenty-four arrangements; FIG. 3 shows twelve of them. The interconnection arrangements of all these configurations are described in Table I. Note that for the network deployment disclosed in this document, all twenty-four arrangements are equivalent because they retain the correct path from the transmitter to the receiver in the network. The arrangement shown in the embodiment example corresponds to arrangement 632.

[0031] FIG4 shows embodiment 700, a variation of arrangement 632, in which all ports are grouped in an array with a fixed spacing (e.g., 127 μm or 250 μm). In this embodiment, the waveguides are written into different layers within the glass, with spacing of at least 40 μm to avoid crosstalk. Ports 515, 520, 525, 550, 555, 560, and 565 are partially shown in the figure. For example, eight waveguides of ports 510, 525, 550, and 565 are marked in the figure. From the top view of the figure, it can be seen that there is crossing between waveguides (e.g., waveguide 710). However, as shown in FIG5, when viewed from the top at a 45-degree tilt, all adjacent waveguides maintain a minimum spacing. In FIG5, there may be crossing between waveguides 720; however, a perspective view of the same waveguide from a different angle ( FIG4( a) ) shows that there is no crossing between them.

[0032] 610 612 614 616 618 620 622 624 626 628 630 632 550a 510e 510f 510g 510g 510e 510f 510h 510h 510f 510e 510g 510h 550b 515e 515f 515g 515g 515e 515f 515h 515h 515f 515e 515g 515h 550c 520e 520f 520g 520g 520e 520f 520h 520h 520f 520e 520g 520h 550d 525e 525f 525g 525g 525e 525f 525h 525h 525f 525e 525g 525h 550e 525d 525c 525b 525b 525d 525c 525a 525a 525c 525d 525b 525a 550f 520d 520c 520b 520b 520d 520c 520a 520a 520c 520d 520b 520a 550g 515d 515c 515b 515b 515d 515c 515a 515a 515c 515d 515b 515a 550h 510d 510c 510b 510b 510d 510c 510a 510a 510c 510d 510b 510a 555a 510h 510e 510f 510h 510g 510e 510f 510g 510h 510f 510e 510g 555b 515h 515e 515f 515h 515g 515e 515f 515g 515h 515f 515e 515g 555c 520h 520e 520f 520h 520g 520e 520f 520g 520h 520f 520e 520g 555d 525h 525e 525f 525h 525g 525e 525f 525g 525h 525f 525e 525g 555e 525a 525d 525c 525a 525b 525d 525c 525b 525a 525c 525d 525b 555f 520a 520d 520c 520a 520b 520d 520c 520b 520a 520c 520d 520b 555g 515a 515d 515c 515a 515b 515d 515c 515b 515a 515c 515d 515b 555h 510a 510d 510c 510a 510b 510d 510c 510b 510a 510c 510d 510b 560a 510g 510h 510e 510g 510h 510g 510e 510e 510g 510h 510f 510f 560b 515g 515h 515e 515g 515h 515g 515e 515e 515g 515h 515f 515f 560c 520g 520h 520e 520g 520h 520g 520e 520e 520g 520h 520f 520f 560d 525g 525h 525e 525g 525h 525g 525e 525e 525g 525h 525f 525f 560e 525b 525a 525d 525b 525a 525b 525d 525d 525b 525a 525c 525c 560f 520b 520a 520d 520b 520a 520b 520d 520d 520b 520a 520c 520c 560g 515b 515a 515d 515b 515a 515b 515d 515d 515b 515a 515c 515c 560h 510b 510e 510d 510b 510a 510b 510d 510d 510b 510a 510c 510c 565a 510f 510g 510h 510e 510f 510h 510g 510f 510e 510g 510h 510e 565b 515f 510g 515h 515e 515f 515h 515g 515f 515e 515g 515h 515e 565c 520f 520g 520h 520e 520f 520h 520g 520f 520e 520g 520h 520e 565d 525f 525g 525h 525e 525f 525h 525g 525f 525e 525g 525h 525e 565e 525c 525b 525a 525d 525c 525a 525b 525c 525d 525b 525a 525d 565f 520c 520b 520a 520d 520c 520a 520b 520c 520d 520b 520a 520d 565g 515c 515b 515a 515d 515c 515a 515b 515c 515d 515b 515a 515d 565h 510c 510b 510a 510d 510c 510a 510b 510c 510d 510b 510a 510d

[0033] Table I Interconnection table of module 400

[0034] The preferred embodiments of the present disclosure utilize the DLWW method to write into glass. However, by making some modifications to the wavelength (or wavelengths) of the light, the optics to focus the beam, the repetition rate and / or the power of the femtosecond laser, the disclosed waveguide designs can be written directly onto polymers (enabling flexible DLWW 3D optical circuits) and silicon substrates.

[0035] More advanced techniques may allow for the writing of waveguides using multiple wavelengths or variable reputation rates as the laser is applied through the glass with a focus - this can provide waveguides with variable width or refractive index / contrast (i.e. the difference between the refractive index of the waveguide and the surrounding glass). Dynamically varying the diameter of the laser while maintaining the power density allows one to vary the diameter of the waveguide while keeping the refractive index constant or variable depending on the application. For example, an upper or lower taper can be used for mode conversion, mode matching, mode filtering, stripping modes, or even multimode waveguides.

[0036] FIG6 shows an example of how to use DLWW embodiment 700 or (500) in an assembly or module 800, which is a housing that supports and protects PLC 700, has front and rear openings for optical interfaces (connectors or adapters), and features for organizing fiber optic interconnections. One or more of the faces of the module have multi-port connectors, such as MPO, MMC, or a group of SN, MDC, or a group of duplex connectors (such as LC connectors). In this figure, module 900 has eight MPO ports, which are evenly distributed on the front and back of the device.

[0037] The output of the DLWW device 700 is connected to fiber arrays 830 and 850. Fiber arrays are commonly used to couple light into and out of a PLC and consist of one or two rows of fiber arrays placed on a high precision V-groove chip, which is covered with a cap on top. The end faces of the fiber arrays are optically polished and can be purchased with flat or angled polished end faces with or without anti-reflective coating (ARC). In the disclosed utility model, the fibers from 830 are divided into groups 810, 815, 820, and 825. Similarly, from the other side of the DLWW device 700, the fiber groups from 850 are divided into groups 860, 865, 870, and 875.

[0038] Fiber groups 810, 815, 820, and 825 are connected to fiber groups 910, 915, 920, and 925, respectively, using fusion or mechanical splices indicated by 790. Element 790 may also be a matched pair of multi-fiber connectors in other designs with less insertion loss constraints. Fibers 910, 915, 920, and 925 are terminated in multi-fiber connectors or duplex connector groups placed at the front or rear of module 800.

[0039] Similarly, fiber groups 850, 855, 860 and 865 are connected to fiber groups 950, 955, 960 and 965, respectively, using fusion or mechanical splices indicated at 790. Fibers 950, 955, 960 and 965 terminate in multi-fiber connectors or duplex connector sets located on the rear or front of module 800.

[0040] Modules and DLWW devices with several ports, fiber groups 850, 855, 860 and 865 are connected to fiber groups 950, 955, 960 and 965 respectively using fusion or mechanical splices represented by 790. Optical fibers 950, 955, 960 and 965 are terminated in multi-fiber connectors or duplex connector groups located on the back or front of module 800.

[0041] Module 800 combines a simple grid that can support up to four trunks and four leaf switches. Larger grids can be implemented using the methods shown in RS25595 and DLWW device 700. For example, as shown in FIG. 7 and described in more detail in RS25595, a two-step grid combined by combining the grid applied to a group of optical fibers and then applied to two parts of a single optical fiber increases the mixing of fiber channels within each module. This simplifies the deployment of the network because a significant portion of the network complexity is transferred from the structured wiring switching network to one or more modules 400. Table II shows the grid configuration of module 400, indicating that any one of the sixteen parallel ports is connected to all other parallel ports using a duplex channel. For example, in the first column of the table, port 440 is shown to be connected to 420a, 425a, 430a, and 435a. By grouping several modules 400 and following simple rules to connect a group of uplinks or downlinks horizontally or vertically, the installation becomes cleaner and cable management is greatly improved. For example, FIG. 8 shows a stack of four modules 400, each with four photonic devices 700. Part (a) of the figure shows one side of the module connected to the leaf switch. For simplicity, we mark it as the front. Part (b) shows the opposite side of the same module 400, the back side, which is connected to the backbone switch. Assume that sixteen leaf switches need to be connected to the structure shown in part (d) of the figure, each leaf switch has four MPO uplinks. In this illustrative example, it is assumed that sixteen leaf switches with four uplinks need to be connected to the switching network, and the connection method is described as follows. The uplinks of each leaf switch are connected horizontally in groups of four until the last port of each module 400 is used. For example, 1110 and 1112, the first and last fourth ports of the first module 400 are connected to the uplink ports of the leaf switches L1 and L4, respectively. The uplink of the fifth leaf switch fills port 1114 of the second module 400. This method continues until the uplink of the last leaf switch is connected to port 1116. The backbone ports are allocated on the back of the stacking module 400. For example, if separate spine switches are used, then 1120, 1122, and 1124 correspond to the ports of the first, second, and sixteenth spine switches, labeled S1, S2, and S16, respectively.

[0042] 440 442 444 446 448 450 452 454 456 458 460 462 464 466 468 470 420a 420a 420a 420a 420b 420b 420b 420b 420c 420c 420c 420c 420d 420d 420d 420d 425a 425a 425a 425a 425b 425b 425b 425b 425c 425c 425c 425c 425d 425d 425d 425d 430a 430a 430a 430a 430b 430b 430b 430b 430c 430c 430c 430c 430d 430d 430d 430d 435a 435a 435a 435a 435b 435b 435b 435b 435c 435c 435c 435c 435d 435d 435d 435d

[0043] Table II Grid Configuration of Module 400

[0044] Further example embodiments:

[0045] Example 1 includes an apparatus having multiple optical duplex and parallel connector adapters, some of which connect to network devices in a network and other adapters connect to servers or processing units, wherein the apparatus combines internal photonic circuitry with a mesh in which the optical paths of each transmitter and receiver are matched to provide appropriate optical connections from a transmitting optical fiber to a receiving optical fiber.

[0046] Example 2 includes the apparatus of Example 1 having a plurality of optical duplex and parallel connector adapters, wherein the adapters include an MPO connector and an LC adapter.

[0047] Example 3 includes the apparatus of Example 1 or 2 having a plurality of optical duplex and parallel connector adapters, wherein the apparatus can be stacked to provide various radix Clos network topologies.

[0048] Example 4 includes the apparatus of any of Examples 1-3 having a plurality of optical duplex and parallel connector adapters, wherein the apparatus enables a network to have different levels of oversubscription from 1:1 to 1:12.

[0049] Example 5 includes the apparatus of any of Examples 1-4 having a plurality of optical duplex and parallel connector adapters, wherein the apparatus can be used to scale an optical network from tens of servers to tens of thousands of servers.

[0050] Example 6 includes the apparatus of any of Examples 1-5 having multiple optical duplex and parallel connector adapters, wherein the apparatus provides redundant paths to reduce the risk of network failures due to interconnect errors.

[0051] Example 7 includes the apparatus of any of Examples 1-6 having multiple optical duplex and parallel connector adapters, wherein the apparatus has a small form factor enabling stacking of three modules in one RU, thereby allowing stacking of up to 132 modules per rack.

[0052] Example 8 includes the apparatus having multiple optical duplex and parallel connector adapters of any of Examples 1-7, wherein the processing unit comprises a GPU.

[0053] Example 9 includes a multi-port passive photonic circuit chip. The multi-port passive photonic circuit chip includes a plurality of waveguides written at several layers on a glass substrate, wherein the waveguides maintain at least a minimum spacing along all of their tracks, wherein some of the waveguides follow a grid interconnection diagram to connect transmit ports and receive ports of an optical channel, wherein the waveguides have a circular cross-sectional shape with a refractive index contrast in the range of 0.2% to 2%, wherein the device as described in any one of Examples 1-8 incorporates an internal grid in the multi-port passive photonic circuit chip.

[0054] Example 10 includes a structured cable system. The structured cable system includes a stack of modules, wherein each module has a plurality of optical duplex and parallel connector adapters, wherein the modules incorporate an internal grid in a multi-port passive photonic circuit chip, wherein the stack of modules can be used to deploy or expand various Clos network topologies using a reduced number of interconnects, wherein the multi-port passive photonic circuit chip includes a plurality of waveguides written at several layers on a glass substrate, wherein the waveguides maintain at least a minimum spacing along all of their traces, wherein some of the waveguides follow a grid interconnection diagram to connect transmit ports and receive ports of an optical channel, wherein the waveguides have a circular cross-sectional shape with a refractive index contrast in the range of 0.2% to 2%.

[0055] Although the utility model has been described as having a preferred design, the utility model can be further modified within the spirit and scope of the present disclosure. Therefore, the present application is intended to cover any variation, use or modification of the present utility model using its general principles. Further, the present application is intended to cover deviations from the present disclosure within known or customary practices in the relevant field of the present utility model, and it falls within the limitations of the attached claims.

Claims

1. A device having a plurality of optical duplex and parallel connector adapters, characterized in that, Some adapters connect to network devices in the network, while others connect to servers or processing units, where the devices incorporate internal photonic circuits into the mesh, where the optical paths of each transmitter and receiver are matched to provide the appropriate optical connection from the sending fiber to the receiving fiber.

2. The device according to claim 1, characterized in that: The adapter includes an MPO connector and an LC adapter.

3. The device according to claim 1, characterized in that: The devices can be stacked to provide Clos network topologies of various radixes.

4. The device according to claim 1, characterized in that: The device enables the network to have different levels of oversubscription from 1:1 to 1:

12.

5. The device according to claim 1, characterized in that: The device can be used to scale optical networks from tens of servers to tens of thousands of servers.

6. The device according to claim 1, characterized in that: The device provides redundant paths, thereby reducing the risk of network failure due to interconnect errors.

7. The device according to claim 1, characterized in that: The device has a small form factor, enabling three modules to be stacked in one RU, allowing up to 132 modules to be stacked per rack.

8. The device according to claim 1, characterized in that: The processing unit includes a GPU.

9. A multi-port passive photonic circuit chip, characterized in that: The multi-port passive photonic circuit chip comprises a plurality of waveguides written at several layers on a glass substrate, wherein the waveguides maintain at least a minimum spacing along all their tracks, wherein some of the waveguides follow a grid interconnection diagram connecting transmit ports and receive ports of optical channels, wherein the waveguides have a circular cross-sectional shape with a refractive index contrast in the range of 0.2% to 2%, wherein the device as claimed in any one of claims 1-8 incorporates an internal grid in the multi-port passive photonic circuit chip.

10. A structured cable system, characterized in that: The structured cable system includes a stack of modules, wherein each module has a plurality of optical duplex and parallel connector adapters, wherein the modules incorporate an internal grid in a multi-port passive photonic circuit chip, wherein the stack of modules can be used to deploy or expand various Clos network topologies using a smaller number of interconnects, wherein the multi-port passive photonic circuit chip includes a plurality of waveguides written at several layers on a glass substrate, wherein the waveguides maintain at least a minimum spacing along all their tracks, wherein some of the waveguides follow a grid interconnect diagram to connect transmit ports and receive ports of an optical channel, wherein the waveguides have a circular cross-sectional shape with a refractive index contrast in the range of 0.2% to 2%.

Images