Cross point switch with LED IO ports and imaging fiber optic cables

Abstract

Systems and methods include an optical transceiver configured to connect to an optical cable (102) having a plurality of optical fiber cores, the optical transceiver including a plurality of transmitters (108) and a plurality of receivers (106), where the transmitters are connected to a first group of the optical fiber cores in the optical cable and the receivers are connected to a second group of the optical fiber cores in the optical cable. The optical transceiver includes an array of elements, where each element includes a pLED and a photodetector connected to a programmable wire, such that a training algorithm selects which wires become active wires in a data path. The associated circuitry of a driver such as a pLED and a TIA of a PD is also part of each element.

Description

[0001] This application is a divisional application, and its original application is a PCT application with application number PCT / US2022 / 042034 and application date August 30, 2022, and entered the Chinese national phase on October 17, 2023, with application number 202280029089.7, and the name is "Optical transceiver, switch connected to the optical transceiver and system". Technical Field

[0002] The present disclosure relates generally to networking hardware. More specifically, the present disclosure relates to systems and methods for micro-LED / PD devices and selection in optical interconnects over imaging fibers, and systems and methods for crosspoint switches with μLED IO ports and imaging fiber bundles. Background Art

[0003] Ten-meter interconnect (10MI) represents a new approach to short-distance optical transmission. Traditionally, short-distance optical transmission is considered to be a distance of more than 100 meters. As discussed in this article, the new approach redefines short-distance optical transmission, and it can now be considered on a meter scale. In a 10MI transmitter based on a light-emitting diode (LED), an imaging fiber is used for transmission and a photodetector (PD) is used for the receiver. This imaging fiber (bundle) consists of thousands (or tens of thousands) of fiber cores. As the number of fiber cores increases, the cost of the imaging fiber is only slightly higher, so using a single bundle for bidirectional transmission is more advantageous than two separate bundles. In addition, fewer bundles can be used to reduce costs, which also results in fewer bundles that need to be aligned during installation. Known solutions use separate transmit (TX) and receive (RX) fibers or rectangular LED / PD splits. The alternating splits between the TX and RX areas described in the present disclosure and the increased angular and XY (Cartesian) deviation tolerances through the optional combination of output signals have not been previously disclosed or practiced.

[0004] Additionally, companies are considering OCS (optical circuit switches) to reduce cost and power consumption while increasing bandwidth. Many tend to compromise on packet-by-packet addressing for AI clusters because workloads and traffic patterns tend to be persistent. The current practice is to adjust the ratio of resource pools based on specific workloads (CPU, GPU, memory, accelerator, FPGA, and storage pools). The key performance metric is ultra-low latency, which is especially necessary when the CPU / GUP / TPU pool is physically separated from the memory pool. Some approaches use a combination of tunable lasers and passive gratings that route light based on wavelength. Other OCS are typically implemented with micromirror MEMS devices, but reconfiguration is slow.

[0005] As described herein, existing optical switching approaches have several disadvantages which have so far prevented their widespread adoption in industry. Furthermore, none of these optical approaches are compatible with novel micro-light emitting diode (μLED) based optical link technologies due to the use of unusual “imaging” fiber bundles and short wavelength (blue) light sources for transmission. μLEDs combined with imaging fibers offer very low cost, low power optical links. However, it is currently not feasible to perform optical switches on imaging fiber bundles while maintaining their spatial coherence properties. What is needed is a corresponding switching fabric that would also offer a very low cost, low power approach. Furthermore, electrical based switches have the added benefits of signal regeneration and very fast (sub-nanosecond) switching.

[0006] In the context of networking equipment and devices, there is a need to interconnect adjacent devices, for example, interconnect adjacent devices within 10m. Network equipment is moving towards modular, decentralized hardware, and electrical backplanes are being replaced by cabling. That is, electrical backplanes are being replaced by cabling between modules. One of the electrical cabling methods is twinaxial cabling ("Twinax"). It works well and is cost-effective at 100Gb / s, but as the rate increases, power consumption increases, coverage decreases, and installation becomes more difficult. There is a shift to optical interconnects, and while it provides higher rates relative to electrical interconnects, existing methods have higher cost and complexity issues. Summary of the invention

[0007] In one embodiment, an optical transceiver configured to be connected to an optical cable having K (K>>1) optical fiber cores includes: M transmitters (M and K are integers, M<K); and P receivers (P is an integer, P<K), wherein the M transmitters are connected to a first group of the K optical fiber cores in the optical cable, and the P receivers are connected to a second group of the K optical fiber cores in the optical cable. The optical fiber cores are imaging fibers. The M transmitters are each a micro light emitting diode (LED), and the P receivers are photodetectors (PD). The micro light emitting diodes each transmit at least 1 Gb / s. The optical transceiver also includes a transmitter circuit, which is configured to receive an aggregate transmit signal and cause the aggregate signal to be transmitted as a plurality of lower rate transmit signals, each of which is transmitted by one of the M transmitters over a portion of the first group of the K cores; and a receiver circuit, which is configured to receive a plurality of lower rate transmit signals from the P receivers and create an aggregate receive signal based on the plurality of lower rate transmit signals. The aggregate transmit and aggregate receive signals are at least 100 Gb / s. The optical cable includes a plurality of optical fiber cores used as guard bands. The guard bands are located between adjacent M transmitters, in a first group of K optical fiber cores, and between adjacent P receivers, in a second group of K optical fiber cores. The length of the optical cable is 10 meters or less. The first group of K optical fiber cores and the second group of K optical fiber cores are fixed. The first group of K optical fiber cores and the second group of K optical fiber cores are determined during operation based on the optical cable and the associated connection with the optical transceiver.

[0008] The optical transceiver also includes an alignment circuit connected to the M transmitters and the P receivers, wherein the alignment circuit is configured to select a first group of the K fiber cores and a second group of the K fiber cores. The second group of the K fiber cores is determined based on a limiting parameter, the limiting parameter including any one of i) post transimpedance amplifier (TIA) noise and impact on signal-to-noise ratio (SNR), and ii) direct photodiode photocurrent sum limited by photodiode capacitance. Some of the M transmitters and P receivers operate on a plurality of corresponding K fiber cores. Each of the M transmitters and P receivers operates on a plurality of corresponding K fiber cores. The first group of the K fiber cores and the second group of the K fiber cores are each approximately half of the K fiber cores. One embodiment contemplates having the first group of the K fiber cores and the second group of the K fiber cores arranged in a semicircle. In an alternative embodiment, the first group of the K fiber cores and the second group of the K fiber cores are each arranged in a circle, one of which is located in an inner region and the other is located in an outer ring adjacent to the inner region. K≥1000.

[0009] In an embodiment, an optical switch system includes: a plurality of input ports, each input port configured to connect to an input fiber bundle; a plurality of output ports, each output port configured to connect to an output fiber bundle, wherein each of the input fiber bundle and the output fiber bundle includes a cable having K fiber cores, K>>1; and an electrical crosspoint switch connected to the plurality of input ports and the plurality of output ports, wherein the electrical crosspoint switch is configured to connect a given input port to a corresponding output port, including connecting all signals in the input fiber bundle to the corresponding output fiber bundle. Each input port is a photodiode array, each output port is a light emitting diode array, and wherein the plurality of input ports, the plurality of output ports, and the electrical crosspoint switch are co-packaged together. The plurality of input ports are in a photodiode array circuit, the plurality of output ports are in a micro light emitting diode array circuit, the electrical crosspoint switch is in a switch circuit, and the photodiode array circuit is stacked on the switch circuit. The K fiber cores form an imaging fiber. Each input port and each output port includes a plurality of signals that form an aggregate signal. Each output port includes M transmitters, M and K are integers, M < K, and each input port includes P receivers, P is an integer, P < K. In each input port, one of the P receivers can be adapted to select a destination address in the electrical crosspoint switch. In each input port, one of the P receivers can be adapted to a clock phase. The electrical crosspoint switch includes a plurality of input traces connected to each input port and a plurality of output traces connected to each output port. The plurality of input traces are logically positioned horizontally, and the plurality of output traces are logically positioned vertically relative to the plurality of input traces, and wherein the electrical crosspoint switch further includes a plurality of switches at corresponding intersections of the plurality of input traces and the plurality of output traces. The plurality of input ports and the plurality of output ports are short-distance devices, and further include one or more long-distance optical modems connected to one or more of the output ports. The short-distance device is a few-meter modem, and one of the one or more long-distance optical modems is a coherent modem. Each of the plurality of input ports and the plurality of output ports includes a plurality of data channels, and wherein the plurality of data channels form an aggregate signal for switching via the electrical crosspoint switch. Each input port is a photodiode array and each output port is a laser array.

[0010] In another embodiment, a hierarchical electronic crosspoint switch adapted to selectively couple a plurality of input ports and output ports comprises: a plurality of hierarchical electronic crosspoint switches; a plurality of input ports, each input port being configured to connect to an input fiber bundle; and a plurality of output ports, each output port being configured to connect to an output fiber bundle. Each input port is a photodiode array, and each output port is a micro light emitting diode array. Each of the input fiber bundles and the output fiber bundles comprises a cable having an optical fiber core, wherein the optical fiber core is an imaging optical fiber. Each of the plurality of input ports and the plurality of output ports comprises a plurality of data channels, and wherein the plurality of data channels form an aggregate signal for switching via the electrical crosspoint switch. Each of the plurality of hierarchical crosspoint switches is configured for a data channel, wherein one of the hierarchical crosspoint switches may be configured as a clock channel. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] The present disclosure is illustrated and described herein with reference to the various drawings, wherein like reference numerals are used to represent like system components / method steps when appropriate, and wherein:

[0012] Figure 1 Schematic diagram of the physical alignment features of a multi-core fiber and chip.

[0013] Figure 2 Schematic diagram of the possible angular deviation between the micro light-emitting diode (μLED) and the photodetector (PD) semicircle.

[0014] Figure 3 to Figure 4 A stepped transimpedance amplifier (TIA) with an intermediate analog cross-point is shown.

[0015] Figure 5 is a schematic diagram of the training patterns used to illuminate the μLED array.

[0016] Figure 6 is a schematic diagram of an example 3D die stack of chips for receiving and transmitting via an optical cable.

[0017] Figure 7 is a schematic diagram of multi-core optical fibers aligned on an array without being closely packed.

[0018] Figure 8 Schematic diagram of the μLED array projecting dots onto a multi-core optical cable.

[0019] Fig. 9 is a schematic diagram showing the misalignment of the PD array with the optical fiber core of the optical cable.

[0020] Fig.10 is a schematic diagram of an example of a fuse-based implementation with a 4-PD group.

[0021] Fig.11 is a magnified image of a fiber core cluster showing slight variations in size.

[0022] Fig.12 is a side view of multiple μLEDs and PDs aligned with a bundle of optical fiber cores.

[0023] Fig.13 Schematic diagram of μLEDs and PDs distributed on the chip (including alternative segmentation of TX and RX areas).

[0024] Fig.14 is a schematic diagram of an optical link showing VCSEL arrays at both ends of an MMF optical fiber as well as transmitter and receiver circuits.

[0025] Fig.15 yes Fig.14 Schematic diagram of an optical link showing orthogonal speckle on an MMF optical fiber.

[0026] Fig.16 yes Fig.14 Schematic diagram of an optical link showing the bending sensitivity on the MMF optical fiber.

[0027] Fig.17 yes Fig.14 Schematic diagram of an optical link showing ultra-coarse wavelength division multiplexing (WDM) on a μLED array.

[0028] Fig.18 is used for Fig.14 Schematic diagram of the training and forward error correction (FEC) process of the μLED array in an optical link.

[0029] Fig.19 is a schematic diagram showing a method of mode group multiplexing (MGDM).

[0030] Fig. 20 is a schematic diagram of an imaging fiber bundle linked to the optical switching system of the present disclosure.

[0031] Fig.21 is a schematic diagram of a non-blocking crosspoint switch of the present disclosure.

[0032] Fig. 22 is a schematic diagram of an 8×8 crosspoint switch.

[0033] Fig.23 is a schematic diagram of the hierarchical cross-point switch method of the present disclosure.

[0034] Fig.24 FIG. 4 is a schematic diagram of using the optical switch system of the present disclosure to perform media conversion via a plug. DETAILED DESCRIPTION

[0035] In various embodiments, the present disclosure relates to systems and methods for an optical switch system that provides a combination of a μLED array, a PD, an imaging fiber bundle, and a crosspoint switch as a system on a chip module. The system includes a plurality of input ports, each of which is configured to connect to an input fiber bundle. The system also includes a plurality of output ports, each of which is configured to connect to an output fiber bundle, wherein each of the input fiber bundles and the output fiber bundles includes a cable having a plurality of optical fiber cores. An electrical crosspoint switch is connected to the plurality of input ports and the plurality of output ports, wherein the electrical crosspoint switch is configured to connect a given input port to a corresponding output port, including connecting all signals in the input fiber bundle to the corresponding output fiber bundle.

[0036] In various other embodiments, the present disclosure relates to systems and methods for increasing the tolerance of optical transceivers. Various partitioning methods of μLED and PD devices are contemplated herein, which allow XY (Cartesian) and angle insensitivity of fiber alignment. In addition, additional μLED and PD devices can be strategically located near the dividing line (i.e., the boundary between μLED and PD) to allow additional coarse deviation tolerance. Various embodiments include utilizing dark (unluminous) fiber core protection bands between μLED devices to increase the deviation tolerance between μLED, imaging fiber (fiber core) and PD. Various embodiments include minimizing dead zibes between PDs to increase light collection efficiency. The ability to selectively combine signals from multiple PDs greatly improves SNR, thereby improving link budgets. The selection of specific PD groupings is implemented at the time of manufacturing and fixing or during operation. If a specific PD grouping is selected during operation, it should be advantageous to extend to connected fibers, optical switches and / or use external fiber jumpers that may have angles and Cartesian (X, Y) deviations. Various embodiments of the present disclosure also utilize a specific effective process for determining a specific grouping of PDs as a single data channel output. PD membership in a group can depend on limiting parameters. For example, post-TIA summation is limited by TIA noise and its impact on SNR. Direct PD photocurrent summation is limited by PD capacitance and impact on bandwidth. In addition, various embodiments include additional guard bands on areas of the device (optical transceiver) that may be more expensive or more prone to failure.

[0037] Micro LED / PD Devices and Selection

[0038] Physical alignment features between multi-core fiber bundles and chips exist in the imaging industry, and the present invention relies on this physical alignment but applies to data communications applications. Figure 1is a schematic diagram of the physical alignment features of a multi-core optical cable and a chip, and describes how the industry achieves this alignment in the production of interconnect cables. In an embodiment, the length of the optical fiber bundle is 10m or less. Figure 1 A chip (optical transceiver) 100 and a fiber bundle 102 are shown. The chip 100 includes physical alignment features 104 arranged at the corners of the chip 100. The chip is coupled to a plurality of light emitting diodes (μLEDs) (also referred to as emitters) 108 and photodetectors (PDs) (also referred to as receivers) 106 arranged on the surface of the chip 100. The μLEDs 108 and PDs 106 are arranged in a semicircular arrangement that allows the same type of chip to be used at both ends of the connection. The semicircular arrangement requires X-axis and Y-axis alignment, as well as angular alignment. The fiber bundle 102 also includes the same physical alignment features 104 as the imaging fibers. In Figure 1 The multi-core fiber bundle 102 and chip 100 shown in FIG. 1 require additional μLED 108 and PD 106 channels near the split line 110 and closer to the perimeter. The loose alignment tolerances will require some way to selectively define groups of PD 106 during manufacturing, and the association of groups with specific μLED 108 data channels. This is proposed through electronic fuses or transmission gates.

[0039] It should be understood that the μLED 108 of the present invention can be any light emitting device, such as a micro-LED, a vertical cavity surface emitting laser (VCSEL), or any other device known to those skilled in the art. In addition, the photodetector 106 of the present invention can be any light detection device known to those skilled in the art or a device for converting the photon energy of light into an electrical signal.

[0040] Even with physical alignment, there are alignment tolerances in the X and Y (Cartesian) and angular directions. The semicircular segmentation and associated PD selection circuitry are constructed to tolerate significant deviations, thereby reducing manufacturing costs.

[0041] Specifically consider the angle deviation, Figure 2 It is shown how the present invention can tolerate up to 9 degrees of angular deviation with a reasonable number (about 5%) of excess μLEDs and PDs. A training pattern or program can identify which μLEDs need to be disabled. Figure 2Possible angular deviations between the μLED semicircle 212 and the PD semicircle 214 are shown. The figure shows an equal semicircle dedicated to the μLED to PD connection. In this case, the angular deviation results in a small piece of wasted μLED 216 and a small piece of wasted PD 218. Due to near-end reflection crosstalk, there may be additional wasted PD 218 along the μLED boundary. The figure also shows a situation where it is desired to save more expensive or less reliable devices (i.e., μLEDs in this example). The present disclosure provides a guard band 206 for element-free areas on the μLED semicircle 212. This is geometrically constructed by creating chords in the circle that do not pass through the center, thereby resulting in unequal line segments. This reduces wasted μLEDs, but increases wasted PDs. In addition, various embodiments include multiple optical fiber cores used as guard bands in optical cables.

[0042] An example μLED budget could be as follows:

[0043] ●400 μLED data transmission

[0044] ●20 μLED clock transmission

[0045] ● 20 μLEDs separated by nearly half a moon are disabled due to angle deviation

[0046] ● 20 redundant μLEDs to accommodate effective μLED failures

[0047] ●Total 460 μLEDs

[0048] The calculation to determine the number of μLEDs wasted when the guard band is not used is as follows: 2 patches (wasted μLEDs) * (9 degrees / 360 degrees) * 400 μLEDs = 20 unusable μLEDs (5% of total μLEDs). This example demonstrates a 9 degree deviation, but it should be understood that any deviation is expected.

[0049] For all deviations (X-axis, Y-axis and angle), it is advantageous to selectively detect the best PD group and associate the group with a specific μLED channel. To do this, it is first necessary to take advantage of the fact that blue light has a short absorption length in silicon. This enables low-capacitance PDs, which in turn enables high gain in the first TIA (transimpedance amplifier) ​​stage. This high-gain TIA provides enough signal strength to drive up to 7 analog transmission gates and associated crosspoint stubs. The second TIA stage is used to sum a group of up to 7 PDs into a single channel and drive the trigger gate.

[0050] The complementary metal oxide semiconductor (CMOS) transmission line signal speed is about 1.7E8 m / s, and a 10Gbps signal with a 100ps bit period corresponds to about 17mm. The crosspoint switch is expected to be much smaller and can therefore be treated as a lumped element, greatly simplifying the overall design and ensuring signal integrity without termination. The crosspoint switch can equip each TIA with a series resistor (including a through-gate resistance) and act as a voltage adder for up to 7 or more TIAs. In addition, the signal chain can implement an IA (current amplifier) ​​and the switch can act as a current adder.

[0051] Embodiments disclosed herein may include hardware with different operating specifications. In various embodiments, the μLED is suitable for transmitting at least 1 Gb / s per time. The transmitter circuit can be configured to receive an aggregate transmit signal so that the aggregate signal is transmitted as multiple lower rate transmit signals, each of which is transmitted by one of the μLEDs (transmitters) over a portion of the first group of fiber cores. The receiver circuit can also be configured to receive multiple lower rate transmit signals from the PD (receiver) and create an aggregate receive signal based thereon. In an embodiment, the aggregate transmit and aggregate receive signals are at least 100 Gb / s.

[0052] Although the present invention has treated the crosspoint switch as a single large design, it may be necessary to segment it to control the crosstalk effects of parasitic capacitance on the open transmission gates. This is done because overlap is required in the segments to handle the tolerance required for physical variations. For example, a group of PDs close to the boundary between crosspoint segments will be able to drive two switch segments. Additional drive strength can be provided by additional amplifiers.

[0053] Figure 3 and Figure 4 A TIA is shown with an intermediate analog cross-point 300. Figure 3 , each PD 306 is locally integrated with a 1-stage TIA 322. In this example, 2800 PDs 306 are considered - 7 PDs 306 per μLED - so 400 μLEDs. It should be understood that other embodiments may have any number of PDs 306 with any number of μLEDs, and any number of PDs 306 per μLED. Figure 4 An analog crosspoint switch with transmission gates (2800 inputs, 400 outputs), a training monitor 324, a 2-level signal aggregation device 322 (which can be a voltage or current adder), and a flip-flop gate 326 is shown.

[0054] The training algorithm periodically recalibrates the PD groupings as they are used, which accommodates μLED failures, aging, temperature, bending, XY axis tolerances, angular tolerances, and breakouts. Training can also be done only during the manufacturing phase, which can accommodate breakouts and initial tolerances. Training requires turning on the μLEDs in separate groups and measuring the resulting signal strength through the PD array at the other end of the fiber. An efficient way to search this space is to illuminate the μLED array using a Hadamard pattern ( Figure 5 64 such patterns are shown in FIG. 1 ). This efficiency is very important if it is done to reduce overhead bandwidth at the time of use, and to reduce the cost of adjustment time during the manufacturing process.

[0055] Figure 6 6 shows a cross-section of an example 3D die stack 628. The die stack 628 includes a plurality of fiber cores 630, Figure 1 608 and PD 606. In addition, a plurality of lenses 632 oriented above the plurality of μLEDs 608 and PD 606 are arranged on the chip 100. Likewise, in the present embodiment, the chip 100 includes separate portions 100a and 100b, and the μLEDs 608 and PD 606 are arranged on the separate portions 100a and 100b. In various embodiments, these separate portions are configured as semicircular portions disclosed herein, while other embodiments provide other shapes and configurations of the portions 100a and 100b on which the μLEDs 608 and PD 606 are arranged. Figure 6 It shows how the PD 606 and the stage 1 TIA 622 do not compete for area with the crosspoint switch 634. Although this embodiment shows the first stage TIA 622 competing for area with the PD 606 in a planar CMOS process, it is also considered that the TIA is located under the backlight PD on the same die so that it does not steal light collection capacity from the PD.

[0056] For the cable tap configuration with a single fiber bundle, it is assumed that the same chip 100 is used at both ends for volume and cost reasons. It is also assumed that the cable is composed of a set of fiber cores 630, and no specific orientation or alignment is required between the fiber cores. The training algorithm detects the position of the fiber subgroup on each array, which means that precise alignment is not required.

[0057] Figure 7 This is a schematic diagram of a multi-core cable that is aligned on an array (chip) without precise alignment. Figure 7In the illustrated embodiment, 20 multi-core cables 740 are aligned over a plurality of μLEDs 708 disposed on a chip 100 (transmitting chip), wherein four separate groups of five multi-core cables 740 transmit to four different receiving (RX) chips 742. More specifically, the groups of cables 740 transmit light from the μLEDs 708 of the first chip 100 to the PDs 706 of the receiving chips 742. It can be seen that the cables 740 do not need to be oriented and precisely aligned at the receiving chips 742.

[0058] It should be appreciated that in other embodiments, any number of cables 740 and receiving chips 742 are contemplated, and the semicircular configuration of μLEDs 708 and PDs 706 can be configured in any manner. Figure 7 The present embodiment shown should be interpreted as a non-limiting example.

[0059] Figure 7 Also shown is a transmit (TX) chip 100 and multiple receive (RX) chips 742 both using one chip style. This again reduces cost, requiring only a single type of chip for connectivity. Other embodiments contemplated herein utilize different configurations and orientations of the μLED and PD portions arranged on the chip, with some embodiments requiring different configurations of the TX and RX locations.

[0060] In this embodiment, it is assumed that up to 7 PDs are selected within the crosspoint switch to drive a single channel. This number of PDs collects almost all of the light transmitted from the μLED, which reduces losses and increases coverage. It should be noted that link performance increases when the composite signal-to-noise ratio (SNR) increases. Assume that the electrical SNR of the highest optical power P0 PD is the baseline SNR0~P0 2 / N, where N is the TIA noise power. Then, adding the additional signal P1 will produce the sum of the signal amplitude and the noise power to give the required SNR greater than the baseline SNR.

[0061]

[0062] When the additional signal amplitude is larger than a fraction of the baseline, the inequality holds:

[0063]

[0064] If the third signal is added with power P2, a positive contribution to the SNR occurs when the following equation is satisfied:

[0065]

[0066] Extending to T total TIAs, the equation is:

[0067]

[0068] In the process of deciding whether to add a TIA, all TIAs in the group are first ranked in descending order and the benefit of adding each TIA is calculated using the generalized equation above. Addition of consecutively ranked TIAs is allowed as long as the general criteria above are met and is terminated immediately if the general criteria above are violated. This procedure can be performed during initial training and during operation when the power input to the TIA is changed.

[0069] Figure 8 The μLEDs are shown projecting spots onto a multi-core optical cable 840, which is shown as a fiber bundle 802 (shown as a honeycomb array). The hexagonal honeycomb array represents a single optical fiber 830. Multiple μLED illumination zones 844 and μLED exclusion zones 846 are shown. The μLED exclusion zone 846 shown in this embodiment is at least 2 optical fibers 830 wide, and the μLED illumination zone 844 is 1 or 2 optical fibers 830 wide (2 optical fibers wide in this example). The fiber bundle 802 constitutes the optical cable 840 and is assumed to be a fixed reference point, where the μLED portion of the chip 100 can have arbitrary X-axis, Y-axis and angular deviations. In this embodiment, the diameter of the optical fiber core 830 is 7um and the cladding 848 is 1um. The angular deviation is most severe near the perimeter, as shown in A below.

[0070] ΔA=D*sin(Δθ)~450*sin(2°)~14um

[0071] It should be appreciated that other embodiments may include any size, shape, and orientation of μLED illumination zones 844 and μLED exclusion zones 846. Additionally, other embodiments may include fiber cores 830 of different sizes and shapes and claddings 848 that are larger or smaller. Figure 8 The shown embodiments are to be construed as non-limiting examples.

[0072] Fig. 9 is a diagram showing the PD array 950 and Figure 8 The figure shows the misalignment of the optical fiber core 930. Figure 8 A honeycomb structure of individual PDs 906 on the fiber cores 930 and illumination zones 944 to simulate the connection of an optical cable to a chip housing the PD array 950. Although some losses due to uncollected light were still observed, the group of 4 PDs 906 was able to collect a significant amount of light from most of the fiber cores 930. The group of 7 PDs 906 forming a flower pattern collected almost all of the light but required more complex circuitry. Changing the μLED illumination to 1 core improved the 4 PD collection efficiency.

[0073] This article also considers an implementation with a crosspoint that has fuses instead of transmission gates. In this case, because the PD current is added before the noise-generating TIA, the choice of connected PDs is different. Therefore, combining PDs is advantageous even at lower photocurrents. The limitation in this case is primarily due to the additional capacitive loading from each connected PD, which reduces bandwidth and increases noise, a factor that needs to be considered when combining PDs. Fig.10 is a schematic diagram of an example of a fuse-based implementation with a 4-PD group.

[0074] exist Fig.10 , multiple PD contacts 1052 and PD aggregation into a TIA 1054 are shown. In this embodiment, each PD has potential connections to 4 TIAs, and each PD is allowed to connect to 1 TIA, with the other connections disconnected as indicated by disconnected connections 1056. To establish the best grouping, all PD connections are initially retained, 1 μLED is enabled in continuous wave (CW) mode, and the TIA with the largest CW signal is determined. The connection of this TIA to the surrounding 4 PDs is retained, while the other 3 connections from these PDs to other TIAs are disconnected.

[0075] Fig.10 A magnified image is also provided in to better illustrate the retained and broken connections. The retained connections 1058 can be seen to travel from the TIA 1054 to the surrounding 4 PDs 1052. The broken connections 1056 from the 4 PDs to other surrounding TIAs can also be more clearly seen. The disclosed alignment circuitry connected to the μLEDs and PDs can also be configured to select a set of fiber cores.

[0076] The optical fiber cores discussed in this disclosure are very regular and consistent in shape, although they are not completely regular and may vary slightly in size and shape. In addition, the optical fiber cores discussed herein may be imaging optical fibers known to those skilled in the art. Fig.11 is a magnified image of the fiber core cluster 1130 showing slight variations in size. Fig.11 The total diameter of the fiber core bundle shown in FIG. 1 is 0.35 mm, with a total of 3500 fiber cores 1130. The diameter of a single fiber core 1130 is 5 microns, and the thickness of the cladding 1148 is 1 micron. As can be seen from the figure, the diameter of the fiber core 1130 ranges from 4.5 microns to 4.8 microns, while the diameter of the cladding ranges from 0.8 microns to 0.9 microns. Therefore, the alignment of the μLED spot with the fiber core is different.

[0077] Fig.12is a side view of a plurality of μLEDs 1208 and PDs 1206 aligned with a bundle 1202 of optical fiber cores 1230. The figure shows the TX and RX sides of the connection made by the optical fiber bundle 1202. The connection via the optical fiber bundle 1202 transmits light from the plurality of μLEDs 1208 to the plurality of PDs 1206. Fig.12 Example side views of three illumination alignments (1266a, 1266b, 1266c) are shown, illustrating how the spot size (illumination area) 1244 at the PD 1206 varies with alignment. Fig.12 Also shown in FIG. 1 is a selectively disabled PD 1206. It shows how the number of PDs 1206 used in a channel varies according to the natural alignment of the fiber core 1230.

[0078] A first PD alignment 1260a of PD 1206 is shown as an example alignment with only 1 disabled PD 1262. A second alignment 1260b of PD 1206 shows multiple disabled PDs 1262, where the disabled PDs 1262 are selected based on positioning and light absorption from μLED 1208. In the figure, μLED 1208 transmits light through multiple lenses 1232 and produces light spots 1264 with a spacing approximately equal to 2 fiber cores 1230. As previously described, fiber bundle 1202 includes cladding 1248, which produces loss between fiber cores 1230. Light travels through fiber cores 1230 and exits onto PD 1206, thereby producing RX light spots (illumination areas) 1244. Additional losses are encountered between PDs 1206, where disabled PDs 1262 are selected based on the location where light is not present. When two adjacent fiber cores 1230 emit light onto PD 1206, resulting in some light overlap, an overlapping illumination region 1244 occurs. In addition, a front view depicting a 9-core distribution and a 16-core distribution is shown. The 9-core distribution includes μLEDs 1208 at every 3 fiber cores 1230, while the 16-core distribution includes μLEDs 1208 at every 4 fiber cores 1220.

[0079] Several embodiments herein describe 9 fiber cores 1230 for each μLED 1208. Allowing at least 2 dark cores between μLEDs helps eliminate crosstalk and bias. It should be understood that more dark cores may be present for additional isolation (i.e., 16-core distribution or other), or if limited by component size. Additionally, the present disclosure is not limited to μLEDs and visible blue light. Embodiments of the present disclosure include operation at longer or shorter wavelengths (e.g., 850 nm) known to those skilled in the art.

[0080] Fig.13is a schematic diagram of μLEDs 1308 and PDs 1306 distributed on chip 100, including alternative partitioning of TX (μLED) and RX (PD) regions. In some embodiments, alternative partitioning of the transmit (TX) and receive (RX) regions is contemplated, where each region corresponds to approximately half the area of ​​the chip. An alternative annular pattern is described herein that includes the advantage of resisting angular deviations assuming satisfactory X-axis and Y-axis alignment. The figure includes a first chip 100a and a second chip 100b, where the first chip 100b includes μLEDs 1308 along an inner region and PDs 1306 along an outer ring, while the second chip 100b includes μLEDs 1308 along an outer ring and PDs 1306 along an inner region. The first chip 100a and the second chip 100b show two sides of the connection, where the μLED 1308 from the first chip will transmit light to the PD 1306 of the second chip 100b through the core of the optical cable, and the μLED 1308 of the second chip transmits light to the PD 1306 of the first chip 100a.

[0081] In various embodiments, different numbers of TX and RX regions are contemplated. For example, a chip may include any number of TX (μLED) regions and any number of RX (PD) regions. In addition, any combination of TX and RX regions is contemplated herein, such as a number of TX regions that is different from the number of RX regions. The embodiments disclosed herein showing one TX region and one RX region should be interpreted as non-limiting examples.

[0082] Likewise, the present disclosure provides various features for increasing the deviation tolerance of the optical transceiver described herein. The annular division of the μLED and PD devices described herein allows for angular insensitivity of the fiber alignment. In addition, additional μLED and PD devices can be strategically located near the dividing line (i.e., the boundary between the μLED and the PD) to allow for additional coarse deviation tolerance. Various embodiments include utilizing dark (unluminous) fiber core protection bands between the μLED devices to increase the deviation tolerance between the μLED, the imaging fiber (fiber core), and the PD. Various embodiments include minimizing the dark area between the PDs to increase the light collection efficiency. The ability to selectively combine signals from multiple PDs greatly improves the SNR, thereby improving the link budget. The selection of a specific PD grouping is implemented at the time of manufacturing and fixing or during operation. If a specific PD grouping is selected during operation, it should be convenient to expand to the use of external fiber jumpers that connect optical fibers and / or may have angles and Cartesian (X, Y) deviations. The various embodiments of the present disclosure also utilize a specific effective process for determining a specific grouping of PDs as a single data channel output. PD membership in a grouping can depend on limiting parameters. For example, post-TIA summation is limited by TIA noise and its impact on SNR. Direct PD photocurrent summation is limited by PD capacitance and impact on bandwidth. Additionally, various embodiments include additional guard bands on areas of the device (optical transceiver) that may be more expensive or more prone to failure.

[0083] Network environment

[0084] Table 1 below provides background on the interconnect cabling market. The present disclosure focuses on addressing modules and technologies for 10m and smaller to avoid the cost burden of longer distances (eg, 300m). Notably, the inventors believe that there is a need for high bandwidth interconnects for 10m and smaller.

[0085]

[0086] Table 1

[0087] Table 2 below provides context for existing approaches and costs relevant to the present disclosure. This example assumes a 400Gb / s interconnect, but the present disclosure also contemplates 800Gb / s and higher, including 1Tb / s and higher.

[0088]

[0089]

[0090] Table 2

[0091] Optical Link

[0092] Fig.14is a schematic diagram of an optical link 10 showing VCSEL arrays 12a, 12b at both ends of an MMF optical fiber 14 and transmitter circuitry 16 and receiver circuitry 18. Fig.15 is a schematic diagram of the optical link 10 showing orthogonal speckles on the MMF optical fiber 14. Fig.16 is a schematic diagram of the optical link 10 illustrating bend sensitivity on the MMF optical fiber 14 . Fig.17 1 is a schematic diagram of an optical link 10 showing ultra-coarse wavelength division multiplexing (WDM) on a VCSEL array 12a. To simplify the illustration, the optical link 10 is shown as a unidirectional configuration. Those skilled in the art will recognize that actual applications will include a bidirectional configuration with another set of devices. For example, the VCSEL array 12a, the PD array 12b, the transmitter circuit 16, and the receiver circuit 18 can be integrated in a single form factor, such as a module, circuit, etc.

[0093] The transmitter circuit 16 includes a transmit multiple-input multiple-output (MIMO) digital signal processor (DSP) connected to a digital-to-analog converter (DAC), which is connected to the VCSEL array 12a. In one embodiment, the VCSEL array 12a is a 14×14 array with a total of 196 pixels, supporting 10 Gb / s per pixel. With 100 active pixels, it supports 1 Tb / s and is approximately 140 μm×140 μm in size.

[0094] The MMF fiber 14 can be about 10 m of a 125 μm graded index MMF (GRIN MMF). A 62.5 μm MMF supports 220 SDM channels. The larger 125 μm MMF increases speckle, which makes sorting easier and allows for larger VCSELs. Dispersion is not an issue at 10 m and low baud. The VCSEL array 12 a is configured to drive the GRIN MMF fiber 14. The CSEL array 12 a is larger than the MMF input face.

[0095] The VCSEL array 12a can be an RGB VCSEL array, whereby VCSELs of different colors are placed closer together. The VCSEL array 12b is a sensor without RGB passive color filters. This is because the speckle pattern is orthogonal to sufficiently different wavelengths. The VCSEL arrays 12a, 12b may be on-chip integrated devices.

[0096] The training algorithm determines which VCSELs are able to couple light into the MMF fiber 14 and which are not. This avoids precise manufacturing alignment requirements. The continuous training algorithm detects dynamic physical disturbances (e.g., bends, temperature, vibration) in the MMF fiber 14 and recalibrates the transmission matrix. This can also be used to detect physical tampering of high security systems, detect seismic activity, detect cable movement by installers, etc.

[0097] The MMF fiber 14 is connected to the VCSEL array 12b, which can include a 20x20 sensor array with 400 pixels. The receiver circuit 18 includes gain and analog-to-digital converters (ADCs) and a receiver DSP.

[0098] The present disclosure includes a low symbol rate that avoids inter-symbol interference (ISI) problems due to modal dispersion and chromatic dispersion at distances <10m. This applies even to blue wavelengths around 500nm.

[0099] Advantageously, the optical link 10 can be constructed with current consumer technology, ie, VCSEL arrays 12a, 12b with integrated lenses, sensor arrays. The present disclosure exploits various orthogonal dimensions: amplitude / phase / frequency / color / space, to achieve high capacity at low cost.

[0100] The present disclosure also contemplates other types of MMF 14, such as large diameter (1000 μm) multimode GRIN POF (Plastic Optical Fiber), such as OM-GIGA.

[0101] The present disclosure also contemplates single-fiber bidirectional operation without a beam splitter by integrating μLEDs and sensors on the same array.

[0102] The present disclosure may include multiple FMF (few-mode fiber) fan-out cables (optical to optical repeater-splitter).

[0103] Training and Forward Error Correction (FEC)

[0104] Fig.18 is a schematic diagram of the training and forward error correction (FEC) process for the VCSEL arrays 12a, 12b.

[0105] Mode Group Diversity Multiplexing (MGDM)

[0106] The VCSEL pixels are separated sufficiently to drive separate pattern groups, thereby generating separate SDM channels. The received patterns are decorrelated to recover the data. This is Mode Group Diversity Multiplexing (MGDM), which Fig.19 is shown in more detail in .

[0107] Classifier

[0108] One limitation of the proposed SDM concept lies in the number of channels that the MMF fiber 14 can support. To quantify this, the minimum required spatial separation of the optical inputs at the fiber entrance face is considered. Each input can be said to occupy the area of ​​a sphere with a diameter equal to this minimum spacing, which is about 4 in the experiment. The close packing of these equal spheres yields a maximum packing density η of just over 90%. A input is the area occupied by each input, A fiber is the area of ​​the MMF core, so we can calculate the maximum number of inputs N that can operate as parallel SDM channels. We find that N ≤ ηA fiber / A input , for an MMF with a core diameter of 62.5, this results in N ≤ 220. However, in this case the number of channels supported by this SDM approach is more likely to be limited by the effectiveness of the mode classification method at the receiver.

[0109] A correlation-based classifier must separate non-zero cross-correlations down to 1 / sqrt(N) from zero-mean cross-correlations. This becomes more difficult for larger N. A partial solution would be to reduce the statistical noise that distorts these correlation coefficients. This can be achieved by increasing the number of speckle speckles (currently ~300), for example, by using MMFs with larger cores. On the other hand, classification results obtained with linear classifiers indicate that in order to obtain robust operation, the number of speckle intensity samples should exceed the number of SDM channels only by a small margin. Therefore, the number of speckle spots on the fiber endface only needs to be a small fraction more than the number of SDM channels. As a rule of thumb, when fewer SDM channels are needed, fewer fiber modes are also needed to produce the required number of speckles. Therefore, in this classification scheme, it is actually advantageous to use FMFs (with smaller core sizes) compared to MMFs. In general, for large N, classification becomes more difficult.

[0110] The pattern generated by multiple beams has a lower speckle contrast than the pattern generated by any single beam, where the speckle contrast C1 is approximately 1 / sqrt(2) due to polarization diversity. When n lasers are turned on simultaneously, the speckle contrast decreases to C1 / sqrt(n). In general, lower speckle contrast is expected to make the pattern classification task more difficult, because in this case speckle contrast can be regarded as the signal-to-noise ratio.

[0111] The detectors in such an array need only sample the local speckle intensity (rather than imaging the full view).

[0112] Pulse stretching

[0113] For dispersion, the unit is ps / (nm*km), the blue μLED is nominally 500nm, and the MMF is about 100ps / (nm*km). The spectral line width of μLED is 20nm to 100nm, but μLED has been reduced to 5nm. Assume that the spectral width of the example is 10nm.

[0114] 100ps / (nm*km)*10nm*(1 / 500km)=2ps dispersion

[0115] Relative to the symbol period of 500ps, a simple guard band is sufficient.

[0116] For modal dispersion, GRIN fiber reduces modal dispersion. In GRIN fiber, the longer paths spend most of their time in the lower refractive index material with faster speeds. The shortest path is the axial path that spends all of its time in the higher refractive index material and has the slowest speed.

[0117] However, if one simply compares the slowest and fastest paths in the GRIN fiber, this yields a worst-case modal dispersion. There will be a better-case scenario, since each μLED will excite a subset of the MMF modes. Assuming each μLED micro-optic lens collimates the light, the number of excited modes is roughly determined by the ratio of the μLED beam diameter to the fiber endface area. Therefore, this is approximately 100 times less than the total modes in the MMF, and therefore 100 times less than the worst-case MMF modal dispersion.

[0118] A GRIN MMF with a real imperfect profile might have a pulse broadening of about 500ps / km. Therefore, the pulse broadening for 2m of fiber will be 1ps in all modes. This is the minimum guard band inserted into the 500ps symbol period. In practice, the pulse broadening is 100 times smaller because subgroup modes as described above are being excited.

[0119] Therefore, we can take advantage of a static capture of the speckle pattern, as pulse broadening should not be an issue at our symbol rate. This reduces the complexity of the equalizer, and thus reduces product cost.

[0120] Architectural Tradeoffs

[0121] Table 3 illustrates the architectural choices and trade-offs

[0122]

[0123]

[0124]

[0125] Table 3

[0126] Non-data communication applications

[0127] The present disclosure is described with reference to data communications, but those skilled in the art will appreciate that other applications are also contemplated, such as imaging. This may include "endoscopy" such as medical endoscopy, industrial microscopy (sewers, machinery, structures, engine blocks), microscopy, etc. Additionally, this may be used to integrate sensors and displays for in-screen fingerprint sensors. Further still, this may be used in automobiles - cars have many cameras and will increase. Fiber optic bundles enable camera arrays in a compact space: 3D imaging.

[0128] Optical switching system

[0129] In the present disclosure, embodiments provide a novel implementation using a μLED based optical link combined with an electronic cross point switch. This simultaneously achieves low latency, low cost, low power consumption and high bandwidth. The present invention includes μLED and PD (photodetector) IOs and electronic cross point switches all on a single chip (or vertical 3D stack of chips) (see Figures 1 to 13 ). μLED-based optical links provide low-cost and low-power optical interconnects while sacrificing an optical range of less than 10 meters. For simplicity, the present disclosure describes specific embodiments, but it should be understood that other devices may also be considered in other embodiments.

[0130] Fig. 20 is a schematic diagram of an imaging fiber bundle 2302 connected to the optical switching system 2300 of the present disclosure. The fiber bundle 2302 combines hundreds or thousands of individual transmission fiber cores with a thin cladding between each core and a reserved physical position of the core between the input and output faces. The imaging fiber bundles 2302 of this embodiment each contain approximately 4000 cores and transmit 1.6Tb of data, with a typical channel rate of less than 10Gbps (several fiber cores "image" a source to a receiver). In one embodiment, the diameter of the imaging fiber is <1mm, and the entire chip is capable of handling 1250 such bundles, with a total IO capacity of 2000Tb. The fiber bundles 2302 are all unsheathed and grouped together to fall on the photodetector (PD) array 2304 on the faceplate 2308 of the chip. The figure shows a square array, but these arrays can be circular or other shapes known to those skilled in the art. Separate PD arrays 2304 and μLED arrays 2306 are also shown, but they can also be mixed in various shapes (see previous parts of this disclosure).

[0131] In this embodiment, for example, each fiber bundle 2302 has 402 channels carried on 4000 individual fiber cores. Each channel can operate at 4Gbps NRZ, with 1 clock-only channel, 1 address channel, and 400 data-only channels. In addition, the PD array 2304 of this embodiment is capable of supporting 1250 1mm sub-arrays, where a given sub-array is mapped to a single fiber bundle 2302. The μLED array 2306 is similarly capable of supporting 1250 1mm sub-arrays. The electronic crosspoint switch 2310 allows the optical switching system 2300 to switch at the fiber bundle level and is additionally unbuffered, while the trigger 2312 re-times each line. It should be understood that Fig. 20 The embodiments shown in are non-limiting examples, and in other embodiments, the number of optical fiber cores, optical fiber bundles, PD arrays, and μLED arrays may be different.

[0132] Fig.21 2410 includes a plurality of ports 2414, each of which includes a plurality of signals. Port 2414 may be a plurality of data channels that are synchronously grouped together to transmit higher data rate signals that comply with standards (i.e., 100Gb Ethernet, OTU4, ODUFlex, 400Gb Ethernet, etc.). In this embodiment, the crosspoint switch includes 398×398 ports, each of which includes 400 signals, resulting in 64M crosspoints. Each crosspoint includes 2 transmission gate transistors, resulting in a total of 128M transmission gate crystals. Each μLED IO (including the driver) occupies a space of approximately 36×36um, and each bundle uses 400 μLEDs. 398 bundles produce 159.2k μLEDs, occupying approximately 15×15mm. Through control logic, the chip becomes a 50×50mm chip. For power, the μLED is 20mW / 112G, so it is 114W. It should be understood that the present embodiment is a non-limiting example and any combination of the components disclosed herein is contemplated.In addition, each input port and each output port can include multiple signals forming an aggregate signal.

[0133] In various embodiments, the μLED driver and PD TIA are on the same substrate as the crosspoint switch ASIC. The clock signal is transmitted on a separate μLED link and associated with several data channels forming a single port, which makes clock recovery simpler and lower power. One or more ports can be associated with a single fiber bundle. Consider combining the clock and data channels with a crosspoint switch, so that the somewhat randomized association of the received data and clock is compensated by the crosspoint, and the correct input and output mapping is restored by the appropriate clock to provide full 3R signal regeneration (resampling, reshaping, retiming).

[0134] Fig. 22 is a schematic diagram of an 8×8 crosspoint switch. The crosspoint switch 2510 is connected to 8 input ports 2514a and 8 output ports 2514b. The electrical crosspoint switch 2510 further includes a plurality of input traces 2518a connected to each input port 2514a and a plurality of output traces 2518b connected to each output port 2514b. The input port 2514a and the 8 output ports 2514b are suitable for coupling to a plurality of optical fiber bundles (i.e., input optical fiber bundles and output optical fiber bundles). The plurality of input ports 2514a and output ports 2514b can be connected via the plurality of crosspoints 2515 through a plurality of switches 2520. The crosspoint switch 2510 of the present disclosure includes input traces 2518a and output traces 2518b oriented vertically and horizontally relative to each other. In various embodiments, each input port 2514a is a photodiode array and each output port 2514b is a micro light emitting diode array. It should be understood that other arrangements and numbers of input and output lines and ports are also contemplated.

[0135] Fig.23 2510 is a schematic diagram of a hierarchical electrical crosspoint switch method of the present disclosure. The present disclosure provides a hierarchical crosspoint method to simplify control and reduce the size of a single crosspoint. In this example, the fiber bundle has 8 different ports 2514. Each port 2514 can be switched to any output end (i.e., 8 input ends × 8 output ends). However, the data channels and clocks forming the ports are synchronously switched to the same port output end. The clock is used to retime the corresponding data channel outputs. In an embodiment, the multiple input ports 2514a, the multiple output ports 2514b and the electrical crosspoint switch 2510 are co-packaged together.

[0136] While some examples show a single-stage 8×8 port configuration, it is expected that more cross-connect ports will be needed. It is feasible to connect 400 fiber bundles (i.e., a 20×20 arrangement) into a single switch, each carrying 1600Gbps of bandwidth in a 16×100Gbps port arrangement. This is a total of 6400 ports (640Tbps). It is not feasible to build a 6400-port switch as a single entity, but it can be done using the multi-stage approach disclosed in this article.

[0137] The present disclosure provides 32×32 crosspoints for 10Gbps signals, which can be modeled as lumped elements. The switch unit is approximately 10×10 microns in 45nm 12SOI CMOS. Assuming a 32×32 switch, the total signal propagation distance between IO buffers is 640 microns. The speed of the CMOS transmission line in the crosspoint is 1.7e8 m / s, which corresponds to a total propagation delay of 640e-6 / 1.7e8~4ps on the crosspoint in the worst case (excluding buffers). The 10Gbps signal has a bit period of 100ps, so 4ps is not important and the 32×32 crosspoints can be regarded as lumped elements.

[0138] Large switch fabrics can be constructed from smaller, individually buffered and timed cells. A 32×32 crosspoint will occupy approximately 320×320um2. A 3-level reconfigurable non-blocking Clos fabric (m=n=32) will occupy approximately 1mm×10mm and provide 1024 channels. To accommodate 100Gbps ports, 11 channels are required (i.e., the total area for 1024 ports is 11mm×10mm). Clos needs to be expanded to 6144×100Gbps ports, 6 rows and 3 columns (i.e., assuming 45nm 12SOI CMOS, the total area is 66mm×30mm). Current CMOS mask limits are approximately 25mm×30mm, so several individual chips must be integrated using industry-standard multi-chip designs. The total number of cell switches is 19008 cells. Assuming each cell switch consumes 20mW of buffering power, the total power is approximately 400W. It is expected that the optical link will consume 1W / 1Tbps. Therefore, the composite 640Tbps switch optical IO will consume 640W. For a 640Tbps switch with optical IO, the total power consumption is about 1000W, less than 2pJ / bit. In comparison, a typical 400G-DR4 pluggable module is about 18pJ / bit, and low-power CoPackagμLED Optical (CPO) is pursuing an initial design of about 14pJ / bit, neither of which provides any switching functionality.

[0139] Using larger cell switches (e.g., 80×80 instead of 32×32) can significantly reduce real estate and power. Each cell switch will occupy 800×800um2. A 3-stage Clos switch is required (i.e., 11*80*3=2640 cell switches). The switch size is the same as before, about 66mm×30mm, and the power of each switch is 50mW, and the total power is greatly reduced to about 132W.

[0140] The present optical switch system can be controlled by an external controller. A method of addressing using a dedicated μLED is also proposed, which enables a source routing switch. The present embodiment prevents multiple inputs from switching to a single output, however, some amount of multicast (single input to multiple outputs) is possible in the crosspoint design envisioned in various embodiments. The media conversion of current optical switch systems is achieved through plug features. For example, converting from a short distance 10 meter μLED link to a 400 km coherent line. Fig.24 FIG. 4 is a schematic diagram of using the optical switch system of the present disclosure to perform media conversion via a plug.

[0141] In various embodiments, a serialized mode is used. The optical switching system operates on bundles of wires (BoWs) rather than single wires, which are slow and highly parallel buses used to communicate within the chip. By operating at these slow speeds, it allows for very large crosspoint matrices because the resulting stubs do not present signal integrity issues at the slow 4Gbps speeds. It should be understood that other embodiments include other modes (e.g., a serialization step before the crosspoint).

[0142] Additionally, other embodiments utilize switching granularity patterns. One extreme is crosspoint switching for each μLED channel. The other extreme is the system described in the present disclosure (i.e., the switching granularity is at the fiber / port level). Subgroup switching granularity is also considered. The larger the granularity, the more control electronics are required within the crosspoint switch. In various embodiments, lasers are used instead of μLEDs, and packet-based switching can be used by adding more address bits to the dedicated μLED address signal. Additionally, embodiments can utilize time slot guard band clock cycles.

[0143] According to standard practice, multiple switch chips can be connected in parallel to form a larger switch. To achieve this, each chip provides a common clock input / output, which allows other chips to be phase synchronized. In addition, a hybrid switch for short-distance μLED signals and an OCS for long-distance signals is considered. In the hybrid approach, the multiple input ports and the multiple output ports are short-distance devices, and one or more long-distance optical modems connected to one or more of the output ports are also included. The short-distance device can be a ten-meter modem, and the one or more long-distance optical modems can be coherent modems.

[0144] Full 3D monolithic integration of the crosspoint switches of the present disclosure is also contemplated, which allows very short vertical interconnects and thus low capacitance relative to existing 2D tiled structures. Figure 6 , representing a 3D stack of μLED switches. 3D stacking allows the μLED array 100a to be positioned on top of the crosspoint array 634. In various embodiments, the multiple input ports are in the micro-LED array circuit, the multiple output ports are in the photodiode array circuit, and the electrical crosspoint switch is in the switch circuit, and the micro-LED array circuit and the photodiode array circuit are stacked on the switch circuit. This reduces the overall power and increases the scale of the switch. The regular structure of the crosspoint switch facilitates 3D integration; tiles of specific sizes are built by manual layout, and then these tiles are replicated in the X, Y, and Z directions to achieve the most efficient crosspoint switch.

[0145] The optical switch system of the present disclosure provides a combination of μLED arrays, PDs, imaging fiber bundles, and crosspoint switches on a single chip. 3D stacking of μLEDs, PDs, and crosspoint array dies results in a significant increase in density and bandwidth, while using optical cones to concentrate multiple imaging fiber bundles on a single chip. An embodiment provides dedicated μLEDs for clock and address, where the clock and address are shared across multiple μLED data channels forming a port. In addition, BoW switching is used, as opposed to switching serial signals, and IOs can be accessed from the surface of the switch rather than its edge.

[0146] in conclusion

[0147] It should be appreciated that some embodiments described herein may include or utilize one or more general or special purpose processors ("one or more processors"), such as microprocessors; central processing units (CPUs); digital signal processors (DSPs); custom processors, such as network processors (NPs) or network processing units (NPUs), graphics processing units (GPUs), etc.; field programmable gate arrays (FPGAs); and similar unique stored program instructions (including software and firmware) for controlling themselves, so as to implement part, most or all of the functions of the methods and / or systems described herein in combination with certain non-processor circuits. Alternatively, some or all of the functions may be implemented by a state machine without stored program instructions, or in one or more application specific integrated circuits (ASICs), where each function or certain combinations of certain functions are implemented as custom logic or circuits. Of course, combinations of the foregoing methods may be used. For some embodiments described herein, the corresponding devices in hardware and optionally with software, firmware, and combinations thereof can be referred to as "circuits configured to ...", "logic configured to ...", etc., to perform a set of operations, steps, methods, processes, algorithms, functions, techniques, etc. on the digital and / or analog signals of the various embodiments described herein.

[0148] In addition, some embodiments may include a non-transitory computer-readable medium having instructions stored thereon for causing a computer, server, device, apparatus, at least one processor, circuit / circuitry, etc. to perform the functions as described and claimed herein. Examples of such non-transitory computer-readable media include, but are not limited to, hard disks, optical storage devices, magnetic storage devices, read-only memories (ROMs), programmable ROMs (PROMs), erasable PROMs (EPROMs), electrical EPROMs (EEPROMs), flash memory, etc. When stored in a non-transitory computer-readable medium, software can include instructions executable by one or more processors (e.g., any type of programmable circuitry or logic) that, in response to such execution, cause the one or more processors to perform a set of operations, steps, methods, processes, algorithms, functions, techniques, etc. as described herein for various embodiments.

[0149] Although the present disclosure has been illustrated and described herein with reference to preferred embodiments and specific examples thereof, it is apparent to one of ordinary skill in the art that other embodiments and examples may perform similar functions and / or achieve similar results. All such equivalent embodiments and examples are within the spirit and scope of the present disclosure, are contemplated thereby, and are intended to be covered by the appended claims. In addition, it should be noted that the various elements, operations, steps, methods, processes, algorithms, functions, techniques, etc. described herein may be used in any and all combinations with each other.

Claims

1. An optical switching system (2300), comprising: one or more input ports, each input port configured to connect to an input fiber optic bundle (2302); one or more output ports, each output port being configured to connect to an output fiber bundle (2302), wherein each of the input fiber bundle (2302) and the output fiber bundle (2302) comprises K optical fiber cores, K>>1; and An electronic crosspoint switch (2310) connected to the one or more input ports and the one or more output ports, wherein the electronic crosspoint switch (2310) is configured to connect a given input port to a corresponding output port, including connecting a signal in the input fiber bundle (2302) to a corresponding output fiber bundle (2302).

2. The optical switching system (2300) according to claim 1, wherein: Each input port is a photodiode array, and each output port is a micro light emitting diode array.

3. The optical switching system (2300) according to claim 1, wherein: Each input port is a photodiode array and each output port is a micro light emitting diode array, and wherein the one or more input ports, the one or more output ports and the electronic crosspoint switch are co-packaged together.

4. The optical switching system (2300) of claim 1, wherein: The one or more input ports are in a photodiode array circuit, the one or more output ports are in a micro light emitting diode array circuit, and the electronic crosspoint switch is in a switch circuit, and Wherein, the micro light emitting diode array circuit and the photodiode array circuit are stacked on the switch circuit.

5. The optical switching system (2300) according to any one of claims 1 to 4, wherein: The input fiber bundle (2302) and the output fiber bundle (2302) are imaging fibers.

6. The optical switching system (2300) according to any one of claims 1 to 5, wherein: Each input port and each output port includes a plurality of signals forming an aggregate signal.

7. The optical switching system (2300) according to any one of claims 1 to 6, wherein: Each output port includes M transmitters (108), M and K are integers, M<K, and each input port includes P receivers (106), P is an integer, P<K.

8. The optical switching system (2300) of claim 7, wherein: In each input port, one of the P receivers (106) is adapted to select a destination address in the electronic crosspoint switch (2310).

9. The optical switching system (2300) of claim 7, wherein: In each input port, one of the P receivers (106) is adapted for a clock phase.

10. The optical switching system (2300) according to any one of claims 1 to 9, wherein: The electronic crosspoint switch (2310) includes a plurality of input traces (2518a) connected to each of the one or more input ports and a plurality of output traces (2518b) connected to each of the one or more output ports.

11. The optical switching system (2300) of claim 10, wherein: The plurality of input traces (2518a) are logically positioned horizontally, and the plurality of output traces (2518b) are logically positioned vertically relative to the plurality of input traces (2518a), and wherein the electronic crosspoint switch (2310) further comprises A plurality of switches (2520) at corresponding intersections of the plurality of input traces (2518a) and the plurality of output traces (2518b).

12. The optical switching system (2300) according to any one of claims 1 to 11, wherein: The one or more input ports and the one or more output ports are short-range devices, and further comprising One or more long distance optical modems connected to one or more of the one or more output ports and the one or more input ports.

13. The optical switching system (2300) of claim 12, wherein: The short range device is a decameter modem and the one or more long range optical modems are coherent modems.

14. The optical switching system (2300) according to any one of claims 1 to 13, wherein: Each of the one or more input ports and the one or more output ports includes a plurality of data lanes, and wherein the plurality of data lanes form an aggregate signal for switching via the electronic crosspoint switch (2310).

15. The optical switching system (2300) of claim 1, wherein: Each of the one or more input ports is a photodiode array, and each of the one or more output ports is a laser array.

Images