High-speed Interconnects Research Articles

Broadband low-loss polarization-insensitive graphene-integrated photonic crystal fiber modulators in the near-infrared range

The amplitude modulator with the function of converting electric signal to optical amplitude modulation is a key component in high-speed optical communication links. In this paper, we demonstrate a variety of two-dimensional graphene-integrated photonic crystal fiber modulators for efficient near-infrared amplitude modulation with ultra-low insertion loss. Taking the advantages of the perfect total internal reflection of the porous photonic crystal fiber and flexible tunability of the graphene conductivity, the modulation depth of 0.075 dB μm−1 with a low optical loss of 2.13 × 10−4 dB μm−1, a long propagation length of ∼20 mm and polarization-insensitive characteristics in a wide wavelength of 1.3 to 1.8 μm can be achieved. This high-performance modulator may have great potential applications in various high-speed telecommunication, interconnection, and graphene-based integrated photonic systems.

112-GBaud mode-division-multiplexing IM-DD transmission beyond net 6.4 Tb/s over weakly coupled FMF for optical interconnections.

Weakly coupled mode-division-multiplexing (MDM) systems based on intensity modulation and direct detection (IM-DD) are a good candidate for further improving the capacity of short-reach optical interconnections. However, restrained by the modal crosstalk of the transmission link and the reception of degenerate mode groups (DMGs) utilizing bandwidth-limited multimode photodetectors (PDs), high-speed MDM IM-DD has encountered a capacity bottleneck. In this Letter, we investigate a high-speed weakly coupled MDM IM-DD transmission system utilizing a degenerate mode diversity receiver scheme adopting high-bandwidth single-mode PDs over a multiple-ring-core (MRC) few-mode fiber (FMF) and a low-crosstalk mode multiplexer/demultiplexer (MUX/DMUX). An MDM IM-DD transmission with four DMGs and eight wavelengths is experimentally demonstrated with 112-GBaud four-level pulse-amplitude modulation (PAM4) and probabilistically shaped PAM8 per lane over 200-m weakly coupled MRC-FMF. To the best of our knowledge, this is the first experimental demonstration of the MDM IM-DD transmission system with up to 112-GBaud baud rate and beyond 6.4-Tb/s net rate. Meanwhile, the experimental results show that the proposed MDM IM-DD transmission link has a superior performance only adopting a low-complexity feedforward equalizer, making it a promising candidate for high-speed optical interconnections.

ANN/Random forest based performance monitoring in high-speed short-reach optical interconnections

In this work, we have developed signal quality monitoring approaches in 100/400 Gbit/s short-reach transmission systems, with the application of four advanced modulation formats. In 100G and 400G transmission systems, it is shown that accuracies of 100 % have been achieved in the modulation format identification (MFI), with the use of random forest (RF) and multitask learning-based artificial neural network (MTL-ANN) for the four modulation formats mentioned. Meanwhile, average mean-square errors (MSEs) of the monitored optical signal-to-noise ratio (OSNRs) are less than 0.1 dB. Random forest uses up to 29 adders and 190 comparators, reducing its complexity by two orders of magnitude compared to MTL-ANN.

Terminated polymer waveguide with low coupling loss to single-mode fiber for 100 Gbps optical interconnects and beyond.

We designed and fabricated terminated polymer waveguides with low coupling loss to a standard single-mode fiber (SSMF) for 100 Gbps optical interconnects application and beyond. The mode field diameter of the polymer waveguide is designed to perfectly match that of the SSMF with a theoretical coupling loss as small as 0.07 dB. The fabricated polymer waveguides on FR-4 substrates exhibit both a low propagation loss of 0.32 dB/cm and a low coupling loss of less than 0.17 dB to SSMF at a wavelength of 1310 nm. These low loss values indicate good compatibilities with both printed circuit boards and fiber optics. Both single-lane 56 Gbaud (112 Gbps) and 60 Gbaud (120 Gbps) 4-level pulse amplitude modulation (PAM4) optical transmission using terminated polymer waveguides have been successfully demonstrated. The bit error ratio (BER) for PAM4 transmission at both data rates reached a level of 10-4 which is well below the forward error correction (FEC) limitation with a power penalty of less than 1 dB. The bandwidth of the test equipment rather than the polymer waveguide itself limits the high-speed optical interconnect performances in our experiments. The results imply that the polymer waveguide is a promising media for single-lane 100 Gbps optical interconnects application and beyond.

Single wavelength MDM-PDM 800-Gb/s net data rate transmission over a standard multimode fiber employing a Kramers-Kronig receiver.

We propose a high-speed multimode fiber short-reach optical interconnect system based on a Kramers-Kronig (KK) field reconstruction with the mode division multiplexing (MDM) and polarization division multiplexing (PDM) technology. In this work, the LP01, LP21a, LP21b, and LP02 modes are selected as independent channels to carry information. The demonstration achieved the 800 Gb/s net data rate per wavelength with a bit-rate-distance-product (BDP) of 8 Tb/s·km. To the best of our knowledge, this is the highest experimental record of a single wavelength BDP over the SMMF with KK detection. In addition, we discuss the system performance after all multiple-input multiple-output (MIMO) and partial MIMO processing and give guidance on the trade-off between system performance and computational resource.

High-speed carbon nanotube photodetector based on a planarized silicon waveguide

The integration of silicon waveguides with low-dimensional materials with excellent optoelectronic properties can enable compact and highly integrated optical devices with multiple advantages for multiple fields. A carbon nanotube (CNT) photodetector integrated on the silicon waveguide has the potential to meet on-chip high-speed optical interconnection systems, based on the outstanding properties of CNTs such as picosecond-level intrinsic photoresponse time, high charge carrier mobility, broad spectral response, high absorption coefficient, and so on. However, the thermal stability of the device may be compromised due to the local suspension in the channel for the height difference between the WG and the substrate. Here, we report a low-cost and low-optical-loss method to achieve the planarized silicon waveguide. After that, the CNT photodetectors integrated on the original and planarized waveguide with asymmetric palladium (Pd)-hafnium (Hf) metal contacts are fabricated. The influence of this planarization method on the performance of devices is analyzed via comparing the dark leakage current, the leakage current rectification ratio (CRR), the series resistances (R S ), and the photoelectric response. Finally, a CNT photodetector based on the planarized waveguide with a photocurrent (Iph) ∼510.84nA, a photoresponsivity (R I ) ∼51.04mA/W, the dark current ∼0.389µA, as well as a 3 dB bandwidth ∼34GHz at the large reverse voltage −3V is shown.

GLRM: Geometric Layout-Based Resource Management Method on Multiple Field Programmable Gate Array Systems

Multiple field programmable gate array (Multi-FPGA) systems are capable of forming larger and more powerful computing units through high-speed interconnections between chips and are beginning to be widely used by various computing service providers. However, the new computing architecture brings new challenges to the system’s task resource management. Existing resource management methods do not fully exploit resources in Multi-FPGA systems, and it is difficult to support fast resource request and release. In this regard, we propose a geometric layout-based resource management (GLRM) method for Multi-FPGA systems. First, a geometric layout-based task combination algorithm (TCA) was proposed to ensure that the final system can use the available FPGA resources more efficiently. Then, we optimised two resource management algorithms using TCA. Compared with state-of-the-art resource management methods, TCA increases resource flexibility by an average of 6% and resource utilisation by an average of 7%, and the two optimised resource management methods are effective in improving resource management performance.

Parallel wavelength-division-multiplexed signal transmission and dispersion compensation enabled by soliton microcombs and microrings

The proliferation of computation-intensive technologies has led to a significant rise in the number of datacenters, posing challenges for high-speed and power-efficient datacenter interconnects (DCIs). Although inter-DCIs based on intensity modulation and direct detection (IM-DD) along with wavelength-division multiplexing technologies exhibit power-efficient and large-capacity properties, the requirement of multiple laser sources leads to high costs and limited scalability, and the chromatic dispersion (CD) restricts the transmission length of optical signals. Here we propose a scalable on-chip parallel IM-DD data transmission system enabled by a single-soliton Kerr microcomb and a reconfigurable microring resonator-based CD compensator. We experimentally demonstrate an aggregate line rate of 1.68 Tbit/s over a 20-km-long SMF. The extrapolated energy consumption for CD compensation of 40-km-SMFs is ~0.3 pJ/bit, which is calculated as being around 6 times less than that of the commercial 400G-ZR coherent transceivers. Our approach holds significant promise for achieving data rates exceeding 10 terabits.

Fast Prediction Method for Scattering Parameters of Rigid-Flex PCBs Based on ANN.

InGaAs detection systems have been increasingly used in the aerospace field, and due to the high signal-to-noise ratio requirements of short-wave infrared quantitative payloads, there is an urgent need for methods for the rapid and precise evaluation and the optimal design of these systems. The rigid-flex printed circuit board (PCB) is a vital component of InGaAs detectors, as its grid ground plane design parameters impact parasitic capacitance and thus affect weak infrared analog signals. To address the time-intensive and costly nature of design optimization achieved with simulations and experimental measurements, we propose an innovative method based on a neural network to predict the scattering parameters of rigid-flex boards for InGaAs detection links. This is the first study in which the effects of rigid-flex boards on weak infrared detection signals have been considered. We first obtained sufficient samples with software simulation. A backpropagation (BP) neural network prediction model was trained on existing sample sets and then verified on a rigid-flex board used in a crucial aerospace short-wave infrared quantitative mission. The model efficiently and accurately predicted high-speed interconnect scattering parameters under various rigid-flex board grid plane parameter conditions. The prediction error was less than 1% compared with a 3D field solver, indicating an overcoming of the iterative optimization inefficiency and showing improved design quality for InGaAs detection circuits.

Towards Buffer Management with Tiered Main Memory

The scaling of per-GB DRAM cost has slowed down in recent years. Recent research has suggested that adding remote memory to a system can further reduce the overall memory cost while maintaining good performance. Remote memory (i.e., tiered memory), connected to host servers via high-speed interconnect protocols such as RDMA and CXL, is expected to deliver 100x (less than 1µs) lower latency than SSD and be more cost-effective than local DRAM through pooling or adopting cheaper memory technologies. Tiered memory opens up a large number of potential use cases within database systems. But previous work has only explored limited ways of using tiered memory. Our study provides a systematic study for DBMS to build tiered memory buffer management with respect to a wide range of hardware performance characteristics. Specifically, we study five different indexing designs that leverage remote memory in different ways and evaluate them through a wide range of metrics including performance, tiered-memory latency sensitivity, and cost-effectiveness. In addition, we propose a new memory provisioning strategy that allocates an optimal amount of local and remote memory for a given workload. Our evaluations show that while some designs achieve higher performance than others, no design can win in all measured dimensions.

XGNN: Boosting Multi-GPU GNN Training via Global GNN Memory Store

GPUs are commonly utilized to accelerate GNN training, particularly on a multi-GPU server with high-speed interconnects (e.g., NVLink and NVSwitch). However, the rapidly increasing scale of graphs poses a challenge to applying GNN to real-world applications, due to limited GPU memory. This paper presents XGNN, a multi-GPU GNN training system that fully utilizes system memory (e.g., GPU and host memory), as well as high-speed interconnects. The core design of XGNN is the Global GNN Memory Store (GGMS), which abstracts underlying resources to provide a unified memory store for GNN training. It partitions hybrid input data, including graph topological and feature data, across both GPU and host memory. GGMS also provides easy-to-use APIs for GNN applications to access data transparently, forwarding data access requests to the actual physical data partitions automatically. Evaluation on various multi-GPU platforms using three common GNN models with four large-scale datasets shows that XGNN outperforms DGL, Quiver and DGL+C by up to 7.9X (from 2.3X), 15.7X (from 3.3X) and 2.8X (from 1.3X), respectively.

Silicon Multi-band On-chip Mode-division (De)Multiplexing for High-speed Optical Interconnect in Both EDFA and TDFA Wavebands

Silicon Multi-band On-chip Mode-division (De)Multiplexing for High-speed Optical Interconnect in Both EDFA and TDFA Wavebands

Statistical Eye Diagrams for High-Speed Interconnects of Packages: A Review

Statistical Eye Diagrams for High-Speed Interconnects of Packages: A Review

Deep Neural Network-Based Surrogate-Assisted Inverse Optimization for High-Speed Interconnects

Deep Neural Network-Based Surrogate-Assisted Inverse Optimization for High-Speed Interconnects

Polymer-Waveguide-Integrated 2D Semiconductor Heterostructures for Optical Communications.

The demand for high-speed and low-loss interconnects in modern computer architectures is difficult to satisfy by using traditional Si-based electronics. Although optical interconnects offer a promising solution owing to their high bandwidth, low energy dissipation, and high-speed processing, integrating elements such as a light source, detector, and modulator, comprising different materials with optical waveguides, presents many challenges in an integrated platform. Two-dimensional (2D) van der Waals (vdW) semiconductors have attracted considerable attention in vertically stackable optoelectronics and advanced flexible photonics. In this study, optoelectronic components for exciton-based photonic circuits are demonstrated by integrating lithographically patterned poly(methyl methacrylate) (PMMA) waveguides on 2D vdW devices. The excitonic signals generated from the 2D materials by using laser excitation were transmitted through patterned PMMA waveguides. By introducing an external electric field and combining vdW heterostructures, an excitonic switch, phototransistor, and guided-light photovoltaic device on SiO2/Si substrates were demonstrated.

Weighted Decision Enabled Phase Retrieval Receiver With Adaptive Intensity Transformation

Phase-retrieval (PR) receivers can reconstruct complex-valued signals from two de-correlated intensity measurements by the use of iterative PR algorithm, without the assistance of any optical carriers. However, both the computational complexity with hundreds of iterations and the limited PR accuracy prevent it from being applied to high-speed photonic interconnections. In current submission, we propose a weighted decision enhanced PR receiver with adaptive intensity transformation (WD_AIT_PR). The weighted decision based on a compressed sigmoid nonlinear function is helpful to avoid the hard-decision (HD) errors, leading to the performance enhancement of the PR receiver, in terms of both convergence speed and PR accuracy. When the 56 GBaud 16QAM signal after transmission over 80 km standard single mode fiber (SSMF) is reconstructed, the WD_AIT_PR receiver is numerically verified to possess the faster convergency speed with only 70 iterations and the best steady BER performance by reducing the required optical signal-to-noise ratio (OSNR) by 0.8 dB, 1.8 dB and 1.8 dB in order to reach 7% HD forward error correction (HD-FEC), in comparison with the HD_AIT_PR receiver by replacing WD with HD, the PR receiver with the phase reset (PR_PR) and the PR receiver with adaptive intensity transformation (AIT_PR), respectively. Meanwhile, the WD_AIT_PR receiver is robust to transmission impairments, including transmitter-side I/Q imbalance, receiver-side skew, the noise imbalance between two intensity measurements, laser phase noise, and laser wavelength drift. Finally, the enhanced performance of the WD_AIT_PR receiver is experimentally verified through recovering 25 GBaud 16QAM signals under the condition of 40 km SSMF transmission.

Electrical Performance Analysis of High-Speed Interconnection and Power Delivery Network (PDN) in Low-Loss Glass Substrate-Based Interposers.

In this article, electrical performance analysis of high-speed interconnection and power delivery network (PDN) in low-loss glass substrate-based interposers is conducted considering signal integrity (SI) and power integrity (PI). The low-loss glass substrate is a superior alternative to silicon substrate in terms of high-speed signaling and fabrication yield. However, the low-loss of the substrate is vulnerable to power/ground noise in the PDN since the low-loss property of the substrate cannot suppress the noise naturally. In this article, an in-depth electrical performance analysis is conducted based on various measurements and simulations to fully benefit the advantages of the low-loss glass substrate. First, the fabrication process and test vehicles for the analysis are explained. Using the test vehicles, the electrical performance of the glass interposer's high-speed interconnection is compared with those of silicon and organic interposers. The insertion loss, eye-diagrams, and signal bandwidths of three interposer channels are compared and analyzed based on electromagnetic (EM) and circuit simulations. Also, the electrical performance of the through glass via (TGV) channel is measured and compared with through silicon via (TSV) channel. The high-speed interconnection of the glass interposer showed better performance for most of the parameters which is more suitable for maintaining the SI. Even though the low-loss of the glass substrate ensured the SI, power/ground noise issues in the PDN must be analyzed and solved. In this article, various cases inducing the power/ground noise in the PDN are considered, simulated, and measured. To solve the issues, ground TGV design and electromagnetic bandgap (EBG) design are proposed for an efficient broadband suppression of the noise generated in the glass interposer PDN.

Revealing the ultrafast spontaneous emission in plasmon-enhanced monolayer semiconductor nano-light sources

Nanoscale spontaneous light sources are promising alternatives to lasers for high-speed optical communications and interconnections through energy-efficient integrated circuits. Yet, developing the spontaneous light sources faster than lasers is hampered by the detection means (e.g., time-resolved fluorescence spectroscopy). Here, by coupling monolayer WSe2 to individual plasmonic nanocavities, we achieved an efficient spontaneous light source with potential ultrafast modulation bandwidth and superior brightness. The ultrafast radiative decay rates can be determined and derived solely from the experimental parameters by combining the coupling strength and the photoluminescence enhancement in a single nanocavity-WSe2 hybrid. As a result, the hybrid light source has a radiative lifetime down to 350 fs, indicating a potential modulation bandwidth up to 440 GHz, which is 10 times of the traditional semiconductor lasers. Furthermore, the quantum yield is enhanced by a factor of over 300-folds up to 20.8% through making full use of the highly confined nanocavity mode. The nanocavity-WSe2 hybrid we built provides a promising approach for constructing high-speed light-emitting devices.

Two fiber-bundle-type FI/FO devices of low insertion loss and cross talk applied for connecting 13-core 5-mode fiber.

Two fiber-bundle-typed fan-in/fan-out (FI/FO) devices, "wavy hexagon-shaped silhouette" type (W-FI/FO) and "tortoise-shaped silhouette" type (T-FI/FO), have been proposed and manufactured based on tapering glass tubes for docking with a self-made 13-core 5-mode fiber. The W-FI/FO device consists of 19 5-mode fibers and has an extended layout based on the 13-core 5-mode fiber structure. It could dock multiple fibers with 19 or 13 cores of the same size standards. When connecting it with 13-core 5-mode, the average losses (ILs) of its five modes are 1.07 dB, 2.95 dB, 3.42 dB, 3.65 dB, and 4.38 dB. The cross talks of the five linearly polarized (LP) modes are -69.1 dB, -64.7 dB, -44.2 dB, -43.9 dB, and -39.1 dB. The T-FI/FO device has a similar core arrangement to the 13-core 5-mode fiber and its average ILs of the five LP modes are 0.23 dB, 1.31 dB, 2.09 dB, 2.66 dB, and 3.03 dB. The cross talks of its five LP modes between adjacent cores are -72.8 dB, -67.8 dB, -43.6 dB, -40.0 dB, and -35.3 dB. The IL and cross talk of the LP01 mode are of satisfactory values, which are 0.23 dB and -72.8 dB, respectively. These two proposed FI/FO devices are expected to be used for high-speed optical interconnection and fiber communication.

A Statistical Approach for Signal and Power Integrity Co-Design in High-Speed Interconnects Considering Non-Linear Power/Ground Noise and Bit-Patterns.

In this article, a novel statistical approach is proposed and applied to co-design signal and power integrity (SI/PI) in high-speed interconnects considering the non-linear power/ground noise generated by parallel buffers and bit-patterns. With increased data rates and decreased operating voltages, the allowed noise margin in high-speed interconnects is continuously reduced, and this trend requires SI/PI co-design. Specifically, non-linear power/ground noise associated with simultaneous switching circuits sharing a power delivery network (PDN) and bit-patterns must be carefully considered during the interconnects' design and analysis phase. In many cases, conventional electromagnetic (EM) and transient circuit simulators require heavy computational resources or even fail to deliver an accurate result. The proposed statistical method estimates the statistical eye-diagram in the high-speed interconnect considering power/ground noise and bit-patterns such as data bus inversion (DBI) coding. The accuracy and computational efficiency of the proposed method are validated by comparing the result with HSPICE transient simulation result. The proposed method is also compared with conventional statistical methods, such as peak distortion analysis (PDA) and statistical channel simulation in the transient simulator. Lastly, the proposed method is applied to the SI/PI co-design and co-analysis in the high bandwidth memory (HBM) interposer channel. Impacts of decoupling capacitors on hierarchical PDN impedance, statistical eye-diagram of the HBM channel, and bit error rate (BER) Bathtub curves are summarized. Finally, the BER eye-diagram is derived from the estimated statistical eye-diagram for timing and voltage analysis. The impacts of hierarchical PDN design and bit-patterns on SI/PI are discussed.