Optical engine and preparation method thereof
By integrating VCSELs and wavelength division multiplexers on the intermediate layer, and combining high electro-optic coefficient materials and laser direct writing technology, the high speed, high density and low power consumption problems of existing optical interconnect technologies in AI computing clusters are solved, realizing an efficient optoelectronic collaborative architecture suitable for short-distance optical interconnects in AI training clusters.
Patent Information
- Application Number
- CN202610055608.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-01-15
- Publication Date
- 2026-02-13
AI Technical Summary
Existing optical interconnect technologies cannot simultaneously meet the requirements of high speed, high density, low power consumption, and high reliability in ultra-large-scale AI computing clusters. Copper cables are limited by physical characteristics, VCSELs cannot achieve sufficient modulation rates, EMLs are too costly and complex, and silicon photonics solutions have insufficient thermal stability and electrical performance.
Multiple VCSELs are integrated using flip-chip technology, combined with wavelength division multiplexing (WDM) technology. Optical waveguides, WDM multiplexers, and demultiplexers are integrated on an interposer layer. Optical waveguides are fabricated using electro-optic modulation materials with high electro-optic coefficients and laser direct writing technology to achieve efficient transmission and conversion of optical signals. Combined with the tight integration of computing chips, a compact and efficient optoelectronic collaborative architecture is constructed.
It achieves high-speed data transmission capabilities of 1.6Tbps and above, reduces signal latency and energy loss, improves system reliability and vibration resistance, and is suitable for large-scale data exchange needs in AI training processes.
Smart Images

Figure CN121522801A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of semiconductor, in particular to a light engine and a preparation method thereof. BACKGROUND
[0002] With the continuous expansion of AI model scale, the interconnection of super-large-scale computing clusters has become the key to improving AI computing power. The effectiveness of data interconnection directly affects the cluster size, effective computing power rate and availability, making short-distance data interconnection in supercomputing centers a system performance bottleneck. The current mainstream interconnection technologies include copper cable and optical module, the latter of which is divided into VCSEL, EML (Electro-absorption Modulated Laser) and silicon light three technical paths.
[0003] Copper cable technology has low cost, small delay, but the transmission distance is limited (within 10 meters), high-speed signal attenuation is serious, the cable is thick and heavy and difficult to dissipate heat, which is difficult to meet the demand of high-speed transmission above 800Gbps. VCSEL scheme has low power consumption, small size and easy coupling, and is widely used in short-distance communication, but the bandwidth of VCSEL is limited, and the single-channel modulation rate is difficult to break through 200Gbps. The EML scheme has fast modulation speed and stable wavelength, and is suitable for long-distance transmission, but the structure is complex, the cost is high, and it is sensitive to temperature, which is not suitable for short-distance high-density scenarios. Silicon light technology is compatible with CMOS process, but has inherent defects: silicon does not emit light and needs external laser, waveguide loss is large, and the difference in thermal expansion coefficient leads to reliability problems, and the high dielectric constant of silicon substrate easily causes signal crosstalk, which performs poorly in high-density interconnection.
[0004] About 50% of the energy consumption in modern AI training clusters comes from data transmission, and All-to-All communication in a million-card cluster requires high-density and microsecond-level data transmission capability. Existing technologies are difficult to simultaneously meet the requirements of high speed, high density, low power consumption and high reliability: copper cable is limited by physical characteristics; traditional VCSEL cannot achieve sufficient modulation rate; EML has high cost and complexity; and the silicon light scheme has insufficient thermal stability and electrical performance. Therefore, a new type of optical engine architecture is needed, which can achieve a single-channel modulation rate of more than 200Gbps, support an aggregated bandwidth of 1.6Tbps and above, and at the same time have low power consumption and high integration, to meet the stringent requirements of AI computing clusters for short-distance optical interconnection. SUMMARY
[0005] The present application provides an optical engine and a preparation method thereof, which can effectively solve the problems of transmission rate, integration density and the like faced by optical interconnection devices in super-large-scale AI computing clusters in the prior art.
[0006] The first aspect of this application discloses an optical engine comprising: an interposer; at least two VCSELs integrated on the interposer via a flip-chip process for emitting continuous-wave laser signals with different wavelengths; at least two first optical waveguides fabricated on the interposer, each first optical waveguide including an optical input section, an electro-optic modulation section, and an optical output section, wherein the optical input section is optically coupled to the VCSELs to couple the continuous-wave laser signals into the first optical waveguides, the electro-optic modulation section is used to electro-optically modulate the continuous-wave laser signals to generate a first modulated-wave laser signal, and the optical output section is used to output the first modulated-wave laser signal; and a wavelength division multiplexer integrated on the interposer, including at least two input ports and one output port, wherein the input ports of the wavelength division multiplexer are optically coupled to the optical output section of the first optical waveguides to combine the first modulated-wave laser signals with different wavelengths and output them to an output optical fiber optically coupled to the output port of the wavelength division multiplexer.
[0007] According to the optical engine of this application, multiple VCSELs are integrated through flip-chip technology, and wavelength division multiplexing (WDM) technology is used to combine and transmit optical signals of different wavelengths, which significantly improves the data transmission density of a single optical fiber and can achieve high-speed data transmission capabilities of 1.6Tbps and above. Furthermore, the optical waveguide and WDM multiplexer can be fabricated together on the interposer using the same process, resulting in high integration and avoiding optical loss caused by optical coupling between the two.
[0008] The optical engine disclosed in this application further includes: a demultiplexer integrated on the interposer layer, comprising an input port and at least two output ports, wherein the input port of the demultiplexer is optically coupled to an input optical fiber and receives a second modulated laser signal with a different wavelength from the input optical fiber; at least two second optical waveguides fabricated on the interposer layer, wherein one end of the second optical waveguide is optically coupled to the output port of the demultiplexer for transmitting the second modulated laser signal; and at least two photodetectors integrated on the interposer layer and optically coupled to the other end of the second optical waveguides for detecting the second modulated laser signal and outputting a first electrical signal.
[0009] According to the optical engine of this application, a complete optical receiving system is constructed by integrating a demultiplexer, a second optical waveguide, and a photodetector on the intermediate layer, realizing efficient conversion of optical signals to electrical signals. This bidirectional optical communication architecture can process transmitted and received data streams simultaneously in the same optical engine, significantly improving the communication efficiency and integration of the optical engine, and providing full-duplex high-speed optical interconnect capabilities for AI clusters.
[0010] The optical engine disclosed in this application further includes: a computing chip that receives and processes the first electrical signal to generate a second electrical signal; and a substrate for integrating the interposer and the computing chip, wherein the substrate is provided with a circuit for transmitting the second electrical signal to the electro-optic modulation unit to perform electro-optic modulation on the continuous wave laser signal.
[0011] According to the optical engine of this application, by directly integrating the optical engine and the computing chip on the same substrate, the electrical signal transmission path is significantly shortened, reducing signal delay and energy loss. This tightly integrated optoelectronic co-engineering architecture makes the conversion of data between the optical and electrical domains more efficient, which is particularly suitable for the frequent large-scale data exchange needs in AI training, effectively improving the overall computing power utilization of the system.
[0012] In the optical engine disclosed in this application, the continuous wave laser signal is coupled to the first optical waveguide through a vertical grating coupler or a 45° mirror fabricated on the intermediate layer.
[0013] According to the optical engine of this application, the coupling problem between the vertical light output of VCSEL and the planar optical waveguide can be solved by fabricating a vertical grating coupler or a 45° mirror structure on the interposer layer. This optical path reversal design not only improves optical coupling efficiency and reduces optical power loss, but also makes full use of the three-dimensional spatial layout, making the optical engine structure more compact and further improving the integration density. Compared with the traditional fiber optic patch cord coupling method, this monolithically integrated optical coupling structure significantly improves the system's reliability and vibration resistance.
[0014] In the optical engine disclosed in this application, the electro-optic modulation section is a micro-ring modulator or a Mach-Zehnder modulator, and the electro-optic modulation material of the electro-optic modulation section is one of lead zirconate titanate, lithium niobate, or barium titanate.
[0015] According to the optical engine of this application, the modulation frequency is significantly improved by using lead zirconate titanate (PZT), lithium niobate (TFLN), or barium titanate (BTO) with high electro-optic coefficients as electro-optic modulation materials. Among them, lead zirconate titanate (PZT thin film) has a photoelectric coefficient as high as 200 pm / V, approximately six times that of conventional materials, enabling high-speed optical signal modulation with extremely low driving voltage, and a single-channel modulation rate exceeding 200 Gbps. This high-performance modulator not only reduces power consumption but also decreases the physical size of the modulator, making high-density integration possible.
[0016] In the light engine disclosed in this application, the intermediary layer is one of a glass intermediary layer, a sapphire intermediary layer, or a resin intermediary layer.
[0017] In the optical engine disclosed in this application, the intermediary layer is a glass intermediary layer, and the first optical waveguide and the second optical waveguide are fabricated on the intermediary layer by a laser direct writing process.
[0018] According to this application, the optical engine directly fabricates optical waveguides on a glass interposer using a laser direct-writing process, avoiding the complex photolithography and etching steps of traditional silicon photonics processes. This process not only reduces manufacturing costs but also enables precise control over the waveguide's geometry and optical properties, achieving low-loss (<0.1dB / cm) optical signal transmission. Furthermore, laser direct-writing technology exhibits excellent compatibility with glass materials, allowing for the fabrication of more complex three-dimensional optical path structures, providing greater design flexibility for optimizing the optical engine's performance.
[0019] In the optical engine disclosed in this application, the intermediary layer is a glass intermediary layer or a sapphire intermediary layer. A thin film material is prepared on the intermediary layer by one of the following methods: vapor deposition, sol-gel method or thin film transfer method. The first optical waveguide and the second optical waveguide are then processed on the thin film material.
[0020] The optical engine according to this application, by using a glass interposer or a sapphire interposer to replace the traditional silicon interposer, fully utilizes the characteristics of low dielectric loss, high thermal stability and excellent optical performance of glass / sapphire materials, effectively reducing dielectric loss and thermal stress during signal transmission, and improving the working stability of the optical engine under different temperature environments.
[0021] In the optical engine disclosed in this application, the intermediary layer is a resin intermediary layer, and the first optical waveguide and the second optical waveguide are fabricated on the intermediary layer by a nanoimprinting process.
[0022] According to the optical engine of this application, using a resin material, such as optical grade resin, as an intermediary layer, an optical waveguide can be fabricated on the intermediary layer by a nanoimprinting process, which can reduce the fabrication cost while ensuring the optical performance of the optical engine.
[0023] In the optical engine disclosed in this application, the wavelength division multiplexer and the demultiplexer are arrayed waveguide gratings.
[0024] The optical engine according to this application eliminates coupling losses and alignment errors between traditional discrete components by directly fabricating the wavelength division multiplexer and demultiplexer on the glass / sapphire interposer. This integrated design not only significantly reduces the size of the optical engine but also improves the stability and reliability of the optical system. Especially for multi-wavelength wavelength division multiplexing systems around 850nm, this planar waveguide design enables precise control of channel spacing, with inter-channel crosstalk below -30dB, effectively ensuring signal integrity for high-speed data transmission.
[0025] In the optical engine disclosed in this application, the emission wavelength of the VCSEL is from 700nm to 1100nm.
[0026] The optical engine of this application utilizes VCSELs in the 700nm to 1100nm wavelength range as the light source, fully leveraging the advantages of this wavelength range in multimode fiber, such as low transmission loss, high maturity of VCSEL devices, and low cost. This wavelength selection is particularly suitable for short-distance, high-speed data interconnection within 500 meters, highly matching the distance requirements of interconnection within AI training clusters. Simultaneously, VCSEL devices in this wavelength range feature low threshold current, low power consumption, and high modulation rate, perfectly complementing the high-density optical engine architecture of this invention, jointly achieving a low-power, high-bandwidth optical interconnection solution.
[0027] The second aspect of this application discloses a method for fabricating an optical engine, comprising the following steps: Step S1, providing an interposer layer; Step S2, fabricating at least two first optical waveguides and a wavelength division multiplexer on the interposer layer, wherein the first optical waveguide includes an optical input section, an electro-optic modulation section, and an optical output section, and the wavelength division multiplexer includes at least two input ports and one output port; Step S3, integrating at least two VCSELs on the interposer layer using a flip-chip process, wherein the VCSELs emit continuous wave laser signals with different wavelengths; the optical input section is optically coupled to the VCSELs to couple the continuous wave laser signals into the first optical waveguides, the electro-optic modulation section is used to electro-optically modulate the continuous wave laser signals to generate a first modulated wave laser signal, the optical output section is used to output the first modulated wave laser signal, and the input port of the wavelength division multiplexer is optically coupled to the optical output section of the first optical waveguide to combine the first modulated wave laser signals with different wavelengths and output them to an output optical fiber optically coupled to the output port of the wavelength division multiplexer.
[0028] In the fabrication method disclosed in this application, step S2 further includes fabricating a demultiplexer and at least two second optical waveguides on the interposer layer. The demultiplexer includes one input port and at least two output ports. The input port of the demultiplexer is optically coupled to an input optical fiber and receives a second modulated laser signal with a different wavelength from the input optical fiber. One end of the second optical waveguide is optically coupled to the output port of the demultiplexer for transmitting the second modulated laser signal. Step S3 further includes integrating at least two photodetectors on the interposer layer using a flip-chip process. The photodetectors are optically coupled to the other end of the second optical waveguide for detecting the second modulated laser signal and outputting a first electrical signal.
[0029] The preparation method disclosed in this application further includes the following steps: Step S4, providing a computing chip and a substrate, integrating the interposer and the computing chip on the substrate, the computing chip receiving and processing the first electrical signal to generate a second electrical signal, and setting a circuit on the substrate to transmit the second electrical signal to the electro-optic modulation unit to perform electro-optic modulation on the continuous wave laser signal.
[0030] In the fabrication method disclosed in this application, step S2 further includes fabricating a vertical grating coupler or a 45° reflector on the intermediate layer for coupling the continuous wave laser signal to the first optical waveguide. Attached Figure Description
[0031] Figure 1 This is a schematic diagram of the structure of one embodiment of the light engine involved in this application;
[0032] Figure 2 This is a schematic diagram of another embodiment of the light engine involved in this application;
[0033] Figure 3 This is a schematic diagram of another embodiment of the light engine involved in this application. Detailed Implementation
[0034] The present application will be further described below with reference to specific embodiments and accompanying drawings. It is to be understood that the illustrative embodiments of this disclosure are merely for explaining the present application and not for limiting it. Furthermore, for ease of description, the accompanying drawings show only the parts relevant to the present application, and not all of the structures or processes.
[0035] The following specific embodiments illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. Although the description of this application is presented in conjunction with preferred embodiments, this does not mean that the features of this invention are limited to this embodiment. On the contrary, the purpose of describing the invention in conjunction with embodiments is to cover other options or modifications that may be derived based on the claims of this application. To provide a thorough understanding of this application, many specific details will be included in the following description. This application may also be implemented without using these details. Furthermore, to avoid confusion or obscuring the focus of this application, some specific details will be omitted in the description. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other.
[0036] Unless the context otherwise specifies, the terms “contains,” “has,” and “includes” are synonyms. The phrase “A / B” means “A or B.” The phrase “A and / or B” means “(A and B) or (A or B).”
[0037] It should be understood that although terms such as "first," "second," etc., may be used herein to describe various components, units, or data, these components, units, or data should not be limited by these terms. These terms are used merely to distinguish one feature from another. For example, without departing from the scope of the exemplary embodiments, a first feature may be referred to as a second feature, and similarly, a second feature may be referred to as a first feature.
[0038] It should be understood that although directional terms such as "up," "down," "left," and "right" may be used here to describe the positional relationship between the various components, these directional terms are only for the convenience of understanding and are not intended to limit the scope of protection of this application.
[0039] It should be noted that in this specification, similar reference numerals and letters in the accompanying drawings indicate similar items. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.
[0040] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.
[0041] Figure 1 This is a schematic diagram of the structure of one embodiment of the light engine involved in this application. Figure 1 As shown, the optical engine in this embodiment includes an interposer layer 1, at least two VCSELs 2, at least two first optical waveguides 3, and a wavelength division multiplexer 4. The VCSELs 2 are integrated on the interposer layer 1 using a flip-chip process, with their light-emitting surfaces facing the interposer layer 1, emitting continuous wave laser signals with a specific wavelength between 700nm and 1100nm.
[0042] On the interposer layer 1, a 45° reflector (not shown) is fabricated at the light exit path of the VCSEL 2 to deflect the vertically emitted continuous-wave laser signal by 90° and couple it to the light input section 31 of the first optical waveguide 3. The 45° reflector can be fabricated on the interposer layer 1 using laser direct writing to achieve the 45° reflective surface. Alternatively, a 45° bevel can be fabricated on the interposer layer using photolithography and etching processes, and a metal reflective layer can be deposited on the 45° bevel to fabricate the 45° reflector. Besides the reflector, a vertical grating coupler can also be fabricated at the light exit path of the VCSEL 2 on the interposer layer 1 to couple the continuous-wave laser signal to the light input section 31 of the first optical waveguide 3 through diffraction.
[0043] The first optical waveguide 3 includes an optical input section 31, an electro-optic modulation section 32, and an optical output section 33. The electro-optic modulation section 32 adopts a Mach-Zehnder modulator or a micro-ring modulator structure. The electro-optic modulation material can be one of lead zirconate titanate (PZT), lithium niobate (TFLN), or barium titanate (BTO) materials, used for high-speed electro-optic modulation of the continuous wave laser signal to generate a first modulated wave laser signal.
[0044] The wavelength division multiplexer 4 is integrated on the interposer layer 1 and includes at least two input ports and one output port. Its input ports are optically coupled to the optical output section 33 of the first optical waveguide 3, merging first modulated laser signals with different wavelengths (e.g., 830nm, 850nm, 870nm, 890nm) and outputting them to the output optical fiber 5, which is optically coupled to the output port. In this embodiment, high-speed data transmission over a single fiber is achieved using four VCSELs 2 with different wavelengths and corresponding first optical waveguides 3, in conjunction with the wavelength division multiplexer 4. However, this application is not limited to this. For example, according to the optical engine of this application, by selecting appropriate wavelength intervals within the wavelength dynamic range of the VCSEL (700nm to 1100nm), it can also have 6 channels, 8 channels, or more channels. The wavelength division multiplexer 4 can be an arrayed waveguide grating (AWG).
[0045] Intermediate layer 1 can be selected from glass intermediate layer, sapphire intermediate layer or resin intermediate layer.
[0046] When a glass interposer is selected as the interposer layer 1, the glass interposer can be chemically strengthened glass with a thickness of 200μm-500μm, possessing low dielectric constant (3.5-4.5), high thermal conductivity (1.0-1.4W / m·K), and excellent optical transparency. The first optical waveguide 3 and the wavelength division multiplexer 4 can be modified by laser direct writing of the glass interposer using a high-energy femtosecond laser.
[0047] When the intermediate layer 1 is selected as a glass intermediate layer or a sapphire intermediate layer, a thin film material can be prepared on the intermediate layer 1 by one of the following methods: vapor deposition, sol-gel method or thin film transfer method, and the first optical waveguide 3 and wavelength division multiplexer 4 can be fabricated on the thin film material.
[0048] A silicon or silicon nitride layer can be deposited on the intermediate layer 1 using vapor deposition as the core layer. Then, the structure of the first optical waveguide 3 and wavelength division multiplexer 4 is fabricated on the core layer using photolithography and etching processes. Afterward, silicon dioxide or silicon oxynitride is deposited as the upper cladding layer. Alternatively, silicon dioxide or silicon oxynitride can be deposited as the lower cladding layer before depositing the core layer. During the fabrication of the first waveguide 3, the core layer material of the optical waveguide is replaced with lead zirconate titanate (PZT), lithium niobate (TFLN), or barium titanate (BTO) at the electro-optic modulation section 32, and electrodes are fabricated to achieve electro-optic modulation. The first optical waveguide 3 and wavelength division multiplexer 4 can also be fabricated from lead zirconate titanate (PZT), lithium niobate (TFLN), or barium titanate (BTO). Lead zirconate titanate (PZT) films are prepared on the interposer layer 1 using a sol-gel method, or lithium niobate (TFLN) films or barium titanate (BTO) films are prepared using a thin film transfer method (e.g., SmartCut method). Then, ridge waveguides are fabricated on the PZT, TFLN, or BTO films using photolithography and etching processes to realize the structure of the first optical waveguide 3 and the wavelength division multiplexer 4. Electro-optic modulation can be achieved using electrodes fabricated at the electro-optic modulation section 32 of the first optical waveguide 3.
[0049] When using a resin-based interposer 1, epoxy resin, polyimide resin, or polyether acrylate-based resin can be selected to prepare the interposer 1. Then, the first optical waveguide 3 and the wavelength division multiplexer 4 are fabricated on the interposer 1 using a nanoimprinting process. For example, a high-refractive-index, low-adhesion UV adhesive (with a refractive index higher than that of the resin interposer) can be spin-coated onto the resin interposer. Then, the structure of the first optical waveguide and the wavelength division multiplexer is imprinted on the UV adhesive using a template, and finally cured with UV light.
[0050] Figure 2 This is a structural schematic diagram of another embodiment of the light engine involved in this application. Figure 1Based on the illustrated embodiment, this embodiment adds a receiving function module, including a demultiplexer 7, at least two second optical waveguides 8, and at least two photodetectors 9. The demultiplexer 7 includes one input port and at least two output ports. Its input port is optically coupled to the input fiber 6 and is used to receive second modulated laser signals with different wavelengths. One end of the second optical waveguide 8 is optically coupled to the output port of the demultiplexer 7, and the other end is optically coupled to the photodetector 9, used to transmit the separated single-wavelength optical signal. The photodetector 9 is made of InGaAs material, with a response wavelength range of 700nm-1100nm and a bandwidth greater than 50GHz, used to convert the received optical signal into a first electrical signal. This bidirectional communication architecture allows a single optical engine to simultaneously process the transmitted and received data streams, significantly improving communication efficiency. The demultiplexer 7 and the second optical waveguide 8 can be fabricated together using various methods described above for fabricating the first optical waveguide 3 and the wavelength division multiplexer 4.
[0051] Figure 3 This is a schematic diagram of another embodiment of the light engine involved in this application. Figure 2 Based on the illustrated embodiment, this embodiment further integrates a computing chip 10. The computing chip 10 (such as a GPU, TPU, or dedicated AI accelerator) is used to receive and process the first electrical signal output by the photodetector 9 and generate a second electrical signal. The substrate 11 adopts high-density interconnect (HDI) technology to integrate the interposer 1 and the computing chip 10. The computing chip 10 can be integrated on the substrate 11 by flip-chip or wire bonding. The substrate 11 is provided with connection circuitry, which can efficiently transmit the second electrical signal processed by the computing chip 10 to the electro-optic modulation unit 32 through the through-hole (TGV) structure provided on the interposer 1 for electro-optic modulation of the continuous wave laser signal. This optoelectronic co-integration architecture significantly shortens the electrical signal transmission path, reduces signal delay to the nanosecond level, and reduces power consumption by more than 30%, making it particularly suitable for the frequent large-scale data exchange needs during AI training.
[0052] This application also provides a method for preparing a light engine, including the following steps.
[0053] Step S1 involves providing an interposer layer, which can be a glass interposer layer, a sapphire interposer layer, or a resin interposer layer. Specifically, a chemically strengthened glass (e.g., Corning Willow Glass) or c-plane sapphire with a thickness of 200-500 μm can be used as the interposer layer. Alternatively, a photosensitive epoxy resin or a polyether acrylate-based resin can be used. For glass interposers, ion-exchange chemical strengthening treatment can be performed to form a compressive stress layer on the surface, resulting in a surface roughness Ra ≤ 0.5 nm. Glass or sapphire interposers possess low dielectric constants (3.5-4.5), high thermal conductivity (1.0-1.4 W / m·K), and excellent optical transparency, providing an ideal platform for subsequent optoelectronic device integration. Using resin as an interposer layer can reduce manufacturing costs while ensuring optical performance.
[0054] Step S2: At least two first optical waveguides and a wavelength division multiplexer are fabricated on the interposer layer. The first optical waveguides include an optical input section, an electro-optic modulation section, and an optical output section. The wavelength division multiplexer includes at least two input ports and one output port. The wavelength division multiplexer can be an arrayed waveguide grating (AWG).
[0055] Taking laser direct writing as an example, a femtosecond laser direct writing system can be used (e.g., a femtosecond laser with a wavelength of 1030 nm, a pulse width of 300 fs, and a repetition frequency of 1 MHz). This system scans within the interlayer to form a waveguide core layer with a refractive index increment Δn > 0.01 and an arrayed waveguide grating (AWG). The cross-sectional dimensions of the optical waveguide can be 5-20 µm. The channel spacing of the AWG can be 3.2 nm, with an insertion loss of less than 2 dB and crosstalk below -30 dB.
[0056] Alternatively, thin-film deposition can be used. First, the cladding and core layers are sequentially deposited on the intermediate layer using processes such as plasma-enhanced chemical vapor deposition (PECVD) and sputtering. Then, the structure of the first optical waveguide and wavelength division multiplexer is etched using photolithography and etching processes, and finally the cladding is deposited.
[0057] The first optical waveguide and wavelength division multiplexer can also be fabricated from lead zirconate titanate (PZT), lithium niobate (TFLN), or barium titanate (BTO) materials. A lead zirconate titanate (PZT), lithium niobate (TFLN), or barium titanate (BTO) thin film is prepared on the interposer 1, and then a ridge waveguide is fabricated on the thin film material using photolithography and etching processes to realize the structure of the first optical waveguide and wavelength division multiplexer.
[0058] The first optical waveguide and wavelength division multiplexer can also be made of photosensitive epoxy resin or polyether acrylate resin, and fabricated on the interlayer through nanoimprinting.
[0059] Step S3 involves integrating at least two VCSELs onto the interposer using a flip-chip process. The VCSELs have emission wavelengths in the range of 700 nm to 1100 nm, and their emission wavelengths differ from each other. A high-precision die-mount machine can be used for flip-chip bonding, employing gold-tin alloy (AuSn) bumps as the bonding material. Bonding is completed in approximately 300°C via reflow soldering for about 10 seconds. During bonding, it is essential to ensure optical coupling between the VCSEL's output port and the optical input section of the first optical waveguide.
[0060] In the optical engine prepared by the above steps: the optical input section is optically coupled to the VCSEL to couple the continuous wave laser signal emitted by the VCSEL to the first optical waveguide; the electro-optic modulation section is used to electro-optically modulate the continuous wave laser signal to generate a first modulated wave laser signal; the optical output section is used to output the first modulated wave laser signal; the input port of the wavelength division multiplexer is optically coupled to the optical output section of the first optical waveguide to combine the first modulated wave laser signals of different wavelengths and output them to the coupled output optical fiber through its output port.
[0061] Step S2 further includes fabricating a demultiplexer and at least two second optical waveguides on the interposer layer using the same fabrication process as the first optical waveguide and wavelength division multiplexer. The demultiplexer (e.g., AWG) includes one input port and at least two output ports. Its input port is used for optical coupling with the input optical fiber to receive second modulated laser signals of different wavelengths from the input fiber. One end of the second optical waveguide is optically coupled to the output port of the demultiplexer for transmitting the demultiplexed single-wavelength second modulated laser signal.
[0062] Step S3 also includes integrating at least two photodetectors (PDs, such as InGaAs PD arrays) onto the interposer layer using a flip-chip process, either simultaneously with or after VCSEL integration. During flip-chip bonding, it is necessary to ensure that the photosensitive surface of the photodetector is optically coupled to the other end of the second optical waveguide. The photodetector is used to detect the second modulated laser signal transmitted through the second optical waveguide and convert it into a first electrical signal output.
[0063] The fabrication method of the optical engine involved in this application further includes: Step S4, providing a computing chip (such as a GPU, TPU, or dedicated AI accelerator) and a substrate (such as a high-density interconnect substrate, HDI-PCB), and integrating the interposer (i.e., the chip integrating optoelectronic devices) that has completed the above steps together with the computing chip on the substrate. The computing chip is used to receive and process a first electrical signal from a photodetector and generate a second electrical signal. A connection circuit is fabricated on the substrate, which transmits the second electrical signal output by the computing chip to the electro-optic modulation section on the interposer, thereby driving it to electro-optically modulate the continuous wave laser signal from the VCSEL. To achieve efficient electrical interconnection, glass vias (TGV) and redistribution layers (RDL) can be pre-fabricated on the interposer to lead the electrodes of the electro-optic modulator and photodetector to the back side of the interposer, and then connect them to the substrate through bumps. The computing chip can be integrated on the substrate using a flip-chip process.
[0064] For example, a 10μm diameter TGV via can be drilled in the interposer using a femtosecond laser. A 50nm thick titanium (Ti) metal layer and a 200nm thick copper (Cu) metal layer are then sequentially sputtered and deposited on the inner wall of the TGV via as an adhesion layer and a seed layer, respectively. Cu is then electroplated to fill the entire TGV via, and a 10μm linewidth / spacing redistribution layer (RDL) is fabricated. The computing chip and the interposer (glass-based optoelectronic chip) are flip-chip bonded together on an HDI-PCB substrate, interconnected via Cu-Cu hybrid bonding. Epoxy resin can also be filled at the bottom of the overall structure to enhance mechanical strength.
[0065] In addition, before integrating the interposer and computing chip onto the substrate, the interposer wafer needs to be diced into individual dies. Furthermore, the optical end face is polished using chemical mechanical polishing (CMP) to achieve a surface roughness Ra≤5nm, and a 400nm thick SiO2 antireflective film (refractive index 1.46) is deposited on the end face using an ion beam sputtering process to achieve a reflectivity R<0.2%@850nm.
[0066] Step S2 further includes fabricating a vertical grating coupler (VGC) or a 45° mirror on the intermediate layer at the endpoint of the optical input section of the first optical waveguide corresponding to the preset light emission position of the VCSEL. This structure is used to couple the vertically emitted continuous-wave laser signal from the VCSEL to the optical input section of the horizontally extending first optical waveguide.
[0067] Vertical grating couplers (VGCs) can be implemented by forming a surface grating structure with a specific period in the waveguide end region through laser direct writing or photolithography etching.
[0068] A 45° reflector can be fabricated on an interposer layer by laser direct writing to achieve the 45° reflective surface. Alternatively, a 45° bevel can be fabricated on the interposer layer using photolithography and etching processes, and a metal reflective layer can be deposited on the 45° bevel to fabricate the 45° reflector.
[0069] The optical engine disclosed in this application achieves a high degree of functional integration and performance breakthrough through architectural innovation. Its core lies in integrating complete optical transmission, reception, and electrical processing functions onto a single interposer layer, constructing a compact, efficient, and independently operable optoelectronic collaborative unit. Specifically, this optical engine monolithically integrates a multi-wavelength VCSEL array, a high-speed electro-optic modulator, a wavelength division multiplexer / demultiplexer, an optical waveguide, and a photodetector array on the same glass, sapphire, or resin interposer layer, thereby realizing a full-duplex optical communication core with integrated transmission and reception at the physical level. Furthermore, through advanced packaging technology, a computing chip (such as a GPU / TPU) is integrated with this photonic chip on a substrate, forming a tightly coupled "computing-optical interconnect" module. This not only greatly shortens the electrical interconnect distance and significantly reduces signal delay and power consumption, but also makes this module an independent subsystem with signal processing capabilities. This highly integrated design enables each optical engine unit to function as a standardized node, directly exchanging and cascading high-speed, high-bandwidth data with other similar units via optical fiber. This perfectly meets the stringent requirements of ultra-large-scale AI computing clusters for high-density, low-latency, and scalable fully interconnected networks, fundamentally improving system-level computing efficiency and energy efficiency ratio.
[0070] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Where there is no conflict, the embodiments and features described in the embodiments of this application can be combined with each other. Therefore, the scope of protection of this application should be determined by the scope of the claims.
Claims
1. A light engine, characterized in that, include: Intermediate layer; At least two VCSELs are integrated on the interposer layer using a flip-chip process to emit continuous wave laser signals with different wavelengths; At least two first optical waveguides are fabricated on the interposer layer. Each first optical waveguide includes an optical input section, an electro-optic modulation section, and an optical output section. The optical input section is optically coupled to the VCSEL to couple the continuous wave laser signal into the first optical waveguide. The electro-optic modulation section is used to electro-optically modulate the continuous wave laser signal to generate a first modulated wave laser signal. The optical output section is used to output the first modulated wave laser signal. A wavelength division multiplexer, integrated on the interposer layer, includes at least two input ports and one output port. The input ports of the wavelength division multiplexer are optically coupled to the optical output section of the first optical waveguide, combining the first modulated laser signals with different wavelengths and outputting them to the output optical fiber optically coupled to the output port of the wavelength division multiplexer.
2. The light engine according to claim 1, characterized in that, Also includes: A demultiplexer, integrated on the interposer layer, includes one input port and at least two output ports. The input port of the demultiplexer is optically coupled to an input optical fiber and receives a second modulated laser signal with a different wavelength from the input optical fiber. At least two second optical waveguides are fabricated on the interposer layer, and one end of the second optical waveguide is optically coupled to the output port of the demultiplexer for transmitting the second modulated laser signal. At least two photodetectors are integrated on the intermediate layer and optically coupled to the other end of the second optical waveguide to detect the second modulated laser signal and output a first electrical signal.
3. The light engine according to claim 2, characterized in that, Also includes: The computing chip receives and processes the first electrical signal to generate a second electrical signal; A substrate is provided for integrating the interposer layer and the computing chip. The substrate is provided with circuitry for transmitting the second electrical signal to the electro-optic modulation unit to perform electro-optic modulation on the continuous wave laser signal.
4. The light engine according to any one of claims 1 to 3, characterized in that, The continuous wave laser signal is coupled to the first optical waveguide through a vertical grating coupler or a 45° mirror fabricated on the intermediate layer.
5. The light engine according to any one of claims 1 to 3, characterized in that, The electro-optic modulation section is a micro-ring modulator or a Mach-Zehnder modulator, and the electro-optic modulation material of the electro-optic modulation section is one of lead zirconate titanate, lithium niobate, or barium titanate.
6. The light engine according to claim 2 or 3, characterized in that, The interlayer is one of a glass interlayer, a sapphire interlayer, or a resin interlayer.
7. The light engine according to claim 6, characterized in that, The intermediate layer is a glass intermediate layer, and the first optical waveguide and the second optical waveguide are fabricated on the intermediate layer by a laser direct writing process.
8. The light engine according to claim 6, characterized in that, The intermediate layer is a glass intermediate layer or a sapphire intermediate layer. A thin film material is prepared on the intermediate layer by one of the following methods: vapor deposition, sol-gel method or thin film transfer method. The first optical waveguide and the second optical waveguide are then prepared on the thin film material.
9. The light engine according to claim 6, characterized in that, The intermediate layer is a resin intermediate layer, and the first optical waveguide and the second optical waveguide are fabricated on the intermediate layer by a nanoimprinting process.
10. The light engine according to claim 2 or 3, characterized in that, The wavelength division multiplexer and the demultiplexer are arrayed waveguide gratings.
11. The light engine according to any one of claims 1 to 3, characterized in that, The emission wavelength of the VCSEL is from 700 nm to 1100 nm.
12. A method for preparing a light engine, characterized in that, Includes the following steps: Step S1: Provide an intermediary layer; Step S2: Fabricate at least two first optical waveguides and a wavelength division multiplexer on the interposer layer. The first optical waveguide includes an optical input section, an electro-optic modulation section, and an optical output section. The wavelength division multiplexer includes at least two input ports and one output port. Step S3: At least two VCSELs are integrated on the interposer layer using a flip-chip process. The VCSELs emit continuous wave laser signals with different wavelengths. The optical input section is optically coupled to the VCSEL to couple the continuous wave laser signal into the first optical waveguide. The electro-optic modulation section is used to electro-optically modulate the continuous wave laser signal to generate a first modulated wave laser signal. The optical output section is used to output the first modulated wave laser signal. The input port of the wavelength division multiplexer is optically coupled to the optical output section of the first optical waveguide to combine the first modulated wave laser signals with different wavelengths and output them to the output optical fiber optically coupled to the output port of the wavelength division multiplexer.
13. The preparation method according to claim 12, characterized in that, Step S2 further includes processing a demultiplexer and at least two second optical waveguides on the interposer layer. The demultiplexer includes an input port and at least two output ports. The input port of the demultiplexer is optically coupled to an input optical fiber and receives second modulated laser signals with different wavelengths from the input optical fiber. One end of the second optical waveguide is optically coupled to the output port of the demultiplexer for transmitting the second modulated laser signal. Step S3 further includes integrating at least two photodetectors on the interposer layer using a flip-chip process. The photodetectors are optically coupled to the other end of the second optical waveguide to detect the second modulated laser signal and output a first electrical signal.
14. The preparation method according to claim 13, characterized in that, It also includes the following steps: Step S4: A computing chip and a substrate are provided. The interposer and the computing chip are integrated on the substrate. The computing chip receives and processes the first electrical signal to generate a second electrical signal. A circuit is provided on the substrate to transmit the second electrical signal to the electro-optic modulation unit to perform electro-optic modulation on the continuous wave laser signal.
15. The preparation method according to any one of claims 12 to 14, characterized in that, Step S2 further includes fabricating a vertical grating coupler or a 45° reflector on the intermediate layer to couple the continuous wave laser signal to the first optical waveguide.
Citation Information
Patent Citations
High-speed VCSEL array integration method and system based on injection locking technology
CN111181002A
Multi-wavelength laser based on vertical gain cavity
CN120320157A
Interposer and preparation method thereof, and photoelectric co-packaging structure
CN120405841A
Electro-optical package and method of fabrication
US20190333905A1
Cited By
A packaging process method of NPO light engine
CN122292044A