Silicon photonics technology promotes advanced high-performance computing
Silicon photonics technology promotes advanced high-performance computing.
Professor Li Pei-wen's team
National Yang Ming Chiao Tung University Institute of Electronics
-
Silicon photonic technology enhances classical high performance, energy-efficiency computing.
Global data centers and the currently popular artificial intelligence applications (such as ChatGPT, DeepSeek, etc.) urgently require high-efficiency and energy-saving hardware (including processor chips and circuit systems) to handle and compute massive amounts of data. However, the performance of commercially available core processors such as CPUs and GPUs has reached saturation since the mid-2010s, and there has been no further advancement. This is because the long-standing Moore's Law has become untenable, and simply shrinking the feature size of transistors and technology nodes can no longer simultaneously ensure computational efficiency and energy consumption. Short-channel effects have caused a dramatic increase in leakage current (energy consumption) of transistors. Furthermore, in order to increase the density of electronic components, the effort to reduce the diameter of metal wires and the electrical interconnects over long distances has become a major bottleneck for energy consumption and speed delays. Although using multi-core processors can enhance the overall performance of computing systems, the connections between circuit boards (board-to-board) and chips (chip-to-chip) within the processor system are still made using cables or metal wires. The more multi-core processors there are, the more frequent and complex the electrical interconnects used at the interfaces become, which actually detracts from the overall performance of the computing system.
In recent years, silicon photonics technology has become the best booster for promoting efficient and energy-saving computing systems. If the ideal characteristics of optical signals, which allow for fast and zero-energy transmission, can be utilized to assist or replace electrical connections with "optical interconnects," it should be possible to break through or improve the predicament of delay and energy consumption in charge signal transmission. Currently, data centers are already using the first generation of silicon photonics technology to replace cable connections between circuit boards. They use so-called pluggable optical transceiver modules to transmit optical signals via optical fibers, which are then converted into electrical signals and transmitted into switches via metal wires. Although this transmission path, which coexists with optical fibers and metal wires, effectively reduces the amount or length of cables used, it still faces issues such as heat generated by metal wires and the attenuation of electrical signal strength. Therefore, processor chip system manufacturers, such as NVIDIA, AMD, and Qualcomm, have all sought technical support for second-generation or 2.5-generation co-packaged optics (CPO) from chip manufacturers like Intel, TSMC, Samsung, and GlobalFoundries. This means integrating the originally scattered optoelectronic modules (such as optical receivers, optical waveguides, optical modulators, current and voltage amplifiers, driver ICs, switches, etc.) on the printed circuit board (PCB) into a single silicon chip through co-packaging. In this way, the use of wire connections can be significantly reduced, achieving improvements in instruction cycles, bandwidth, and reduced power consumption.
In December 2024, at the IEDM international conference held in San Francisco, TSMC announced the launch of a new advanced silicon photonics and packaging technology – the "Compact Universal Photonic Engine (COUPE)" for data center applications.[1–4]The process design kits (PDK) for individual optical components (germanium photodetectors, micro-ring optical modulators, silicon nitride/silicon optical waveguides, grating couplers and edge couplers for connecting external optical fibers, temperature sensors, etc.) have been prepared for chip designers to use. TSMC has published several papers on silicon photonic engines, demonstrating the integration of electronic integrated circuit chips and silicon photonic circuit chips using CoWoS (Chip-on-Wafer-on-Substrate) packaging technology. TSMC expects to officially launch pluggable optical transceiver modules co-packaged on PCBs in 2025, and in 2026, it will introduce planar integration of photonic integrated circuits and electronic integrated circuits on substrates, as well as vertically-stacking integration of photonic integrated circuits and electronic integrated circuits on interposers.
Although advanced and robust CMOS process technology has successfully produced many advanced electronic processors and integrated circuit systems, it is not easy to implement vertically stacked integrated silicon photonic-electronic integrated circuits on a silicon platform. First, the geometric sizes of optical components in photonic integrated circuits (such as semiconductor lasers, photodetectors, optical modulators, especially optical waveguides, etc.) are far larger than the geometric sizes of electronic components in electronic integrated circuits (such as transistors, capacitors, etc.), with dimensions (such as feature lengths or film thicknesses on the order of sub-mm or mm) primarily determined by the wavelength of the transmitted light. Furthermore, the connections between optical components focus on optical coupling. To reduce coupling loss, the coupling area or coupling length typically needs to be greater than mm squared.2or above millimeters (mm), this is almost a hundred to even a thousand times that of electrical contact in electronic components. However, to reduce light scattering and transmission loss, the sidewall roughness of optical components must be at the nanometer scale (nm-scale). This requirement for sidewall flatness at the nanometer level is not common for electronic components. To realize silicon photonic technology, chip manufacturers such as the predecessor of GlobalFoundries, IBM, began researching silicon photonic technology in the early 2000s (or even earlier) and officially proposed the concept and prototype of silicon photonic optical interconnects in 2008. Intel has long collaborated with the University of California, Santa Barbara (UCSB), Stanford University, and the Massachusetts Institute of Technology (MIT), strategically deploying and deeply rooting silicon photonic optical interconnect technology for many years. In addition to optimizing processes for the epitaxial growth of single crystal thin films, structural design, and optical performance required for individual optical components, they have further developed active optical components and the coupling design and process integration of optical waveguides, even demonstrating optical integrated circuits and systems (e.g., high-density wavelength division multiplexing optical fiber backbone transmission systems). Currently, Intel's leading silicon photonic technology should hold a leading position in the chip manufacturing industry. At present, silicon photonic chips have been integrated into data centers, 5G communications, and other technological fields, opening up huge business opportunities. Companies such as Google, Apple, Meta, Amazon, and Microsoft are also actively developing high-efficiency silicon optical interconnect chip technology, establishing optical connections for short-distance internal data centers (850nm lasers and multimode fibers) and long-distance connections between data centers (1310nm lasers and single-mode fibers). Other European and American companies and research institutions, such as Cisco, STM, and IMEC, AIM Photonics, and IME, have also long been engaged in the technical layout of silicon photonic integrated circuits and CMOS integrated circuits, hoping to further enhance the speed and bandwidth of on-chip signal processing.
Although wafer manufacturers have now launched foundry services for silicon photonic technology, available for optical interconnect applications between circuit boards, to fully leverage the substantial benefits of silicon photonics for efficient and energy-saving computing, on-chip optical interconnects must be initiated. However, current chip foundries are unable to provide directly integrable laser light sources. Professor John Bower from the University of California, Santa Barbara (UCSB) gave a keynote speech at the 2024 IEDM international conference, providing detailed and insightful explanations on the key technologies required to fabricate directly integrable quantum dot laser light sources on silicon platforms, as well as the technological evolution to achieve high-capacity silicon photonic integrated circuits.[5]Generally speaking, semiconductor thin-film laser sources typically require a large driving current to trigger laser emission. However, during long-term operation, as the operating temperature gradually increases, the wavelength of the emitted light is prone to red shift, and the light intensity will decrease accordingly. Therefore, there are often drawbacks such as poor temperature stability, reliability, and lifespan. Furthermore, silicon is an indirect bandgap semiconductor, which has extremely low photoelectric conversion efficiency. Fortunately, in 2008, the Massachusetts Institute of Technology in the United States demonstrated the functionality of germanium semiconductor lasers.[6]By using n-type heavy doping and strain engineering techniques, the indirect bandgap of germanium films can be transformed into a pseudo-direct bandgap. However, the Free Carrier Absorption effect significantly increases the laser threshold (the critical current value, approximately 280 mA/cm.2), resulting in a short operating life for germanium lasers, which is detrimental to the reliability and stability of operating temperatures. Therefore, currently, it can only integrate III-V laser light sources with active/passive components of silicon photonics through packaging or wafer bonding. However, the cost of packaging or wafer bonding is quite high, which is not favorable for market competitiveness.
In contrast, benefiting from the quantum confinement effects in three dimensions, quantum dots effectively trap excitons within them, allowing for a relatively small drive current threshold to activate laser emission, and they can operate stably in higher temperature environments. Furthermore, the dispersed energy levels within the quantum dots help to purify the monochromatic wavelength of quantum dot lasers (with spectral linewidths of about tens of nm or narrower, only 1/10 of the linewidth of conventional laser sources), stabilize pulsed laser mode locking, and extend laser lifetime (by a factor of ten thousand).[5]Moreover, quantum dots possess the quantum physical characteristics of tunable electronic energy structures, which can adjust the emission wavelength by changing the diameter of the sub-dots through formulation. Additionally, with engineering designs such as Surface Plasmon Resonance (SPR), the performance of quantum structure lasers can be further optimized. Professor John Bower specifically introduced his laboratory, where they first etched a silicon dioxide thin film on a silicon platform, exposing the silicon, and then selectively grew InAs quantum dots, demonstrating concrete results of high-quality quantum dot laser light sources.
In addition to quantum dot lasers, germanium photodetectors that can be integrated with silicon nitride/silicon photonic waveguides are also one of the important cornerstones for realizing silicon photonic platforms. Due to the relatively small bandgap energy of germanium semiconductor, approximately 0.66 eV, there are concerns about high dark current, which affects the signal-to-noise ratio of photoconversion, detection capability, and severely causes energy loss. Many studies have pointed out that incorporating quantum dots into the light absorption region structure of photodetectors can reduce dark current, increase the lifetime of photocarriers, thermal stability, and photoreactivity, and can also modulate the detection wavelength of light, as well as miniaturize the thickness/area of the components.[7]As a result, the technological research on quantum dot light sources and photodetectors is also flourishing like mushrooms after rain.
In addition to the manufacturing challenges of components such as quantum dot laser sources and germanium photodetectors, another practical technical challenge for the commercialization of silicon photonics technology is how to accurately and efficiently numerically simulate the pre-designed active and passive components of silicon photonics (e.g., semiconductor lasers, photodetectors, optical modulators, optical waveguides, beam splitters, concentrators, etc.) and assess the overall performance of the photonic integrated circuit system. Professor Jelena Vuckovic from Stanford University, in her keynote speech at the 2024 IEDM International Conference, also demonstrated that her laboratory, through collaborations with other top research institutions, laboratories, and various epitaxial wafer manufacturers, has developed high-speed, high-performance electromagnetic (EM) simulation analysis software and hardware (Stanford Photonics Inverse Design Software, SPINS), and has verified the miniaturized, integrable photonic integrated circuit systems through the mature processes of industry-academia collaboration partners.[8]Through close collaboration between industry and academia in inverse design, the laboratory and its partners have successfully demonstrated scalable quantum and classical silicon photonic integrated optical circuit systems. For example, they can provide broadband multi-channel optical transceiver modules for data center applications, as well as control silicon hole electron spin qubits on a silicon carbide platform using a custom-made miniaturized Ti:Sapphire laser.
2. The Key Role of Silicon Photonic Technology in Quantum Computing
As Professor Jelena Vuckovic said, silicon photonic technology can not only improve the computing speed of traditional classical computing and enhance the transmission efficiency of data centers, but also promote the practical development of qubit technology in quantum computing. In fact, both silicon vacancies and ion trap qubits urgently need silicon photonic integrated circuits to manipulate or read quantum states nearby, in order to truly demonstrate the operation of qubits.
For ion trap qubits that claim to operate at room temperature, most currently use optical table-top light or microwave signals to remotely control ion trap qubits located within a vacuum chamber.[9]When reading the quantum state information of ion trap qubits, it is also focused onto single-photon detectors through optical tabletop high numerical aperture lenses. In a room temperature, vacuum environment, the functionality of ion trap qubits has been verified. However, in terms of increasing the number of qubits and improving the fidelity of actual initialization/control/detection of quantum states, ion trap quantum technology faces many technical challenges.[9]
This is due to the alignment between complex optical components and optical fibers, as well as environmental disturbances such as mechanical vibrations and thermal noise, which generate many additional noise sources that limit the readout fidelity of ion trap qubits. If laser light sources can directly manipulate ion trap qubits through optical waveguides in a vacuum environment, and simultaneously connect single-photon detectors via optical waveguides to directly read the quantum state information of ion trap qubits and output it to CMOS integrated circuits for subsequent signal processing, it would not only eliminate the jitter/drift of optical components in free space and ubiquitous electrical noise but also avoid the problem of alignment between overly long optical fibers and complex optical components. This would be greatly beneficial for "improving" the readout fidelity of ion trap qubits.
3. The technical challenges of applying silicon photonic technology to qubits
Although the research and development of silicon photonic device technology has been ongoing for over thirty years, most of the R&D on silicon photonics has focused on applications such as communications or classical computing. The developed silicon photonic link technology (components such as optical waveguides, optical modulators, optical detectors, and even light sources) emphasizes processing high-speed, high-bandwidth, high-responsivity, or high-power optical/electrical signals for communication applications. In order to manipulate or read out ion trap qubits using silicon photonic devices, it is necessary to handle low-noise, low dark current, or extremely low power (a few photons) near-ultraviolet or visible light signals, and it must also be able to operate stably in low-temperature environments. The following will focus on the application of ion trap qubits, highlighting the key technical requirements and challenges of silicon photonic devices.
Integrated silicon nitride waveguides and gratings
The wavelength range of laser light sources commonly used to control ion trap qubits is approximately 300-2000 nm, covering the near-ultraviolet, visible, and near-infrared spectra. Unfortunately, the silicon photonic waveguides commonly used in current silicon photonic technology have a high absorption rate in the ultraviolet and visible light bands, resulting in significant optical property losses, making them unsuitable for ion trap qubit technology. On the contrary, silicon nitride (Si3N4Optical waveguides are transparent and non-absorbing in the ultraviolet-visible wavelength range, with low optical loss.[10,11]Silicon nitride is a commonly used insulating layer, barrier layer, and protective layer in CMOS process technology, which can be deposited using conventional chemical vapor deposition (CVD) methods. The CVD process formulation can also be fine-tuned according to the actual application requirements to adjust the chemical composition of silicon nitride (e.g.: SixNyEven SiOxNy(Thin film) and refractive index help to modulate the number of modes, light confinement, and transmission loss of silicon nitride waveguides. Since silicon nitride waveguides can cover the visible light (400-1000nm) spectrum that silicon waveguides cannot, they have become the best platform for various experimental chips in recent years, especially in quantum communication/computation.[12]。
2020 ETH Zurich[13]The report integrates ion trap chips with silicon nitride optical waveguides into a single ion trap quantum logic gate. The demonstrated ion trap quantum logic gate uses single-mode optical fibers to inject 729 nm visible light into the silicon nitride optical waveguide, transmitting it to the ion trap chip located in a vacuum, low-temperature environment. This method eliminates the challenges of optical alignment on optical tables, as well as mechanical vibrations and beam point drift, improving the fidelity of the quantum logic gate. However, the ion trap chip demonstrated by ETH Zurich has not yet been integrated with active silicon photonic components such as optical modulators and single-photon detectors. This is because the crystallization nucleation dormancy time of germanium or silicon-germanium on silicon nitride films is very short, making it difficult to grow high-quality single-crystal germanium or silicon-germanium films using selective epitaxial growth methods on silicon nitride thin films, thus preventing further fabrication of silicon photonic active components. Although it is possible to use (1) wafer bonding to bond SOI on the silicon nitride platform and then epitaxially grow the optical active layer – germanium or silicon-germanium films.[14]or imitating (2) STM, IHP company and the University of Toronto to create silicon-germanium modulators and germanium photodetectors on the SOI platform, followed by depositing PECVD silicon nitride film, and after CMP polishing, creating the top layer silicon nitride optical waveguide.[14-16]However, the process proposed for the top-layer silicon nitride waveguide is very difficult to continue executing the high-temperature annealing process for dehydrogenation or densification of silicon nitride, which cannot reduce the number of defects within the silicon nitride waveguide. This action would lead to the relaxation of the lattice of the optical active region at the bottom: silicon-germanium and germanium epitaxial thin films, resulting in the degradation of the performance of the optical active components. To date, on the silicon nitride platform, single-chip integrated germanium/silicon-germanium high-speed optical modulators.[14]High-speed optical detectors[14,15]and laser light source[17]There are very few literature reports. Therefore, it is an important research topic to fabricate optical active components on the silicon nitride platform and to integrate optical active/passive components.
Integrated chip single-photon detector
The quantum state signals of quantum bits are very weak and easily disturbed by noise from the surrounding environment, so there is a need for single-photon detectors that can be directly "built-in" to quickly and accurately read and detect the changes in the number of photons from ion trap quantum bits. Ideally, single-photon detectors should be directly connected to ion trap quantum chips via silicon nitride waveguides, minimizing the crosstalk in "collecting" and "detecting" photons, and further expanding the feasibility of measuring large ion trap quantum bit arrays. Generally, the wavelengths of photons emitted by ion trap quantum bits are mostly between 300–500 nm. However, the currently "most mature" silicon avalanche photodetectors can detect light at 850 nm, which cannot directly detect the states of ion trap quantum bits. Researchers at NIST in the United States have used a custom-built "superconducting" single-photon detector that does not require imaging lenses and cameras, achieving nearly perfect accuracy (with a reading accuracy of over 99.9%) in reading the quantum state of beryllium ions.[18]However, "superconducting" single-photon detectors must operate in an environment close to absolute zero. According to a report from NIST, there is an urgent need for near-ultraviolet silicon-based single-photon detectors that are compatible with CMOS technology to effectively improve detection efficiency and reduce dark count rates. Furthermore, there is a need for single-chip integration of silicon nitride waveguides/gratings with silicon-based single-photon detectors to further reduce coupling losses and noise, as well as to expand the number of ion trap qubits.
Integrated chip light source
In addition to single-photon detectors that can be integrated with ion trap quantum chips, visible light sources coupled with silicon nitride waveguides are also key components for controlling ion trap quantum chips. As Professor John Bower stated in his invited talk at IEDM, achieving integrated light sources on silicon substrates has always been the biggest challenge of silicon photonic technology. Not to mention visible light sources that can be integrated with silicon nitride waveguides.
Literature reports that the use of germanium nanostructures such as quantum wells, quantum wires, and even quantum dots can effectively mitigate the defect issues in the growth of single-crystal germanium films on silicon wafers. Especially, due to the quantum confinement effect, the strong overlapping coupling of electron-hole wave functions within the confined germanium quantum dots greatly enhances the optical transition oscillation strength of the germanium quantum dots, breaking the curse that bulk germanium materials must strictly adhere to energy-momentum (E-k) conservation. Furthermore, by adjusting the diameter of individual germanium quantum dots, the emission energy gap can be tuned to emit light of different wavelengths, thus overcoming the limitation of needing to select different bulk materials to produce light sources of different wavelengths. However, the luminescent volume of a single quantum dot is small and needs to be placed within a resonant cavity. When laser light irradiates the quantum dots/resonant cavity, the 'Purcell effect' rapidly increases the number of optically excited photons within the quantum dots, enhancing the overall quality factor of the luminescence. Commonly used quantum dot resonant cavity structures include photonic crystals, micro-disks, and micro-rings. Given the complexity of the structural design of photonic crystal arrays (such as template thickness, hole diameter, period, and defect modes or modes) and the ultra-high requirements for process precision (which must use advanced lithography systems to expose sub-micron level holes (diameter or period)), the design and fabrication of micron-level micro-disks or micro-rings and resonant cavities are relatively easier, and the emission of in-plane light can couple with adjacent bus waveguides, facilitating integration on the chip.
The microdisk resonator mainly confines the optical field within a disk-shaped optical medium, achieving resonance along the radial direction of the disk at the edge of the micro-ring resonator, generating whispering gallery mode (WGM). The structure of the microdisk resonator offers considerable flexibility and cost advantages in the design and fabrication of electrodes and waveguides. In recent years, research institutions in Europe and the United States have reported embedding various quantum dots (such as silicon, germanium, CdSe) in suspended silicon, germanium, silicon dioxide, or silicon nitride microdisk resonators, demonstrating optically pumped micro lasers. The CNRS-Univ. in France has published a series of papers.[19]On the gallium arsenide substrate, a 300-nm thick n-type layer is first epitaxially grown.+-Germanium, after fabricating suspended germanium microdisks through lithography and etching, silicon nitride is deposited to coat the germanium microdisks, forming a tensile strain n+-Ge active light-emitting layer, demonstrating light-excited germanium micro-lasers. However, the method of using germanium on gallium arsenide (Ge-on-GaAs) is very difficult to transfer to silicon platforms. Tokyo City University proposed a P-I-N germanium quantum dot microdisk diode,[20]It can couple with adjacent waveguides to excite light electrically. However, the vast majority of quantum dot microdisks are fabricated on SOI platforms, which are not suitable for visible light sources. There is an urgent need to develop quantum dot/silicon nitride microdisk visible light sources to smoothly integrate with ion trap quantum chips.
This experimental team reported at the flagship IEDM international conference in 2022 that single-chip integrated silicon nitride waveguides (including: grating couplers and waveguide cones), germanium quantum dot microdisk light sources, and photon detectors can be used for ion trap sensing applications in the near ultraviolet-visible light range, as shown in Figure 1.
__25K12uT49z.jpg "02 (2)")
Fig. 1 Ge quantum dot photodiode and light emitter embedded in silicon nitride
__25K1232Fa4.png)
Fig. 2 Formation of self-organized heterostructures of capping SiO2/Ge QDs within host of Si3N4 on top of SOI as evidenced by TEM, HAADF STEM and EDS maps of elemental Ge (green), N (red), and O (white) micrographs.
After P. W. Li et al., IEDM Tech. Dig. pp. 451-454 (2022).
Reference:
[1] Y. J. Mii, “Semiconductor industry outlook and new technology frontiers,” IEDM Tech. Digest, 1.1, Dec. 2024
[2] S. K. Yeh et al., “Silicon photonics platform for next generation data communication technologies,” IEDM Tech. Digest, 23.3, Dec. 2024
[3] C. H. Fann et al., “Novel parallel digital optical computing system (DOC) for generative AI,” IEDM Tech. Digest, 31.7, Dec. 2024
[4] H. Hsia et al., “EPIC-BOE: An electronic-photonic chiplet integration technology with IC processes for broadband optical engine applications,” IEDM Tech. Digest, 31.8, Dec. 2024
[5] J. Bower, “Integrated quantum dot lasers and high capacity silicon photonic integrated circuits,” IEDM Tech. Digest, 23.1, Dec. 2024
[6] RE Camacho-Aguilera et al., “An electrically pumped germanium laser,” Optics Express, vol. 20, no. 10, pp. 11316-11320, 2012
[7] Hongmei Liu, Fangfang Zhang, Jianqi Zhang, Guojing He, “Performance analysis of quantum dots infrared photodetector,” Proceedings of SPIE, vol. 8193, 81930J (2011)
[8] Jelena Vuckovic, “Scalable quantum and classical photonics,” IEDM Digest, 26.3, Dec. 2024
[9] K. Brown et al., “Materials challenges for trapped-ion quantum computers,” Nat. Rev. Mater. vol. 6, pp. 892–905 (2021).
[10] D. J. Blumenthal et al., “Silicon nitride in silicon photonics,” Proc. IEEE, vol.106, 12 (2018).
[11] P. Munoz et al., “Silicon nitride photonic integration platforms for visible, near-infrared and mid-infrared applications,” Sensors, vol.17, 2088(2017).
[12] A. Orieux et al., “Recent advances on integrated quantum communications,” J. Opt., vol. 18, 083002 (2016).
[13] K. Mehta et al., “Integrated optical multi-ion quantum logic,” Nature vol. 586, pp. 533–537 (2020).
[14] F. Boeuf et al., “A silicon photonics technology for 400 Gbit/s applications,” IEDM Tech. Digest, pp. 775 (2019)
[15] S. Lischke et al., “Silicon nitride waveguide coupled 67GHZ Ge photodiode for non-SOI PIC and EPIC platforms,” IEDM Tech. Digest, pp. 779 (2019)
[16] W. Sacner et al., “Monolithically integrated multilayer silicon nitride-on-Si waveguide platforms,” Proc. IEEE, vol. 16, 2232 (2018)
[17] S. Bao et al., “Low-threshold optically pumped lasing in highly strained germanium nanowires,” Nat Comm. vol. 8, 1845 (2017)
[18] S. L. Todaro et al., “State readout of a trapped ion qubit using a trap-integrated superconducting photon detector” Phys. Rev. Lett. 126, 010501 (2021)
[19] A. Ghrib et al., “Tensile-strained germanium microdisks,” Appl. Phys. vol. 102, 221112 (2013)
[20] X. Xu, T. Maruizumi, and Yasuhiro Shiraki, “Waveguide-integrated microdisk light-emitting diode and photodetector based on Ge quantum dots,” Optics Exp., vol. 22, 3905 (2014)
[21] C. H. Lin, P. Y. Hong, B. J. Lee, H. C. Lin, T. George, and P. W. Li, “Monolithic integration of top Si3N4-waveguided germanium quantum-Dots microdisk light emitters and PIN photodetectors for on-chip ultrafine sensing,” IEDM Tech. Dig. pp. 451454, Dec. 2022.
[22] C. H. Lin, P. Y. Hong, B. J. Lee, H. C. Lin, T. George, and P. W. Li, “Self-organized germanium quantum-dots/Si3N4 enabling monolithic integration of top Si3N4 waveguided microdisk light emitters and PIN photodetectors for on-chip sensing,” IEEE Trans. Electron Dev. vol. 70, no. 4, pp. 2113-2120 (2023)
