The recent surge in AI and ML applications within the semiconductor industry necessitates stringent thermal management, driving the need for highly-efficient two-phase cooling mechanisms. Two-phase heat sinks consisting of parallel microchannels offer significantly higher heat transfer efficiency compared to single-phase liquid cooling. However, flow instabilities and local dryout can be triggered by non-uniform heating and severely degrade the thermal performance causing system failure. Therefore, careful characterization of the heatsinks via spatially resolved temperature measurements is crucial for studying the failure modes, such as dryout, for two-phase heat sinks. In this study, we employ a thermal test vehicle (TTV) comprising of 18 silicon dies integrated on a single interposer, functioning as both the heat source and temperature sensor to characterize thermal performance in a custom two-phase heat sink with high-aspect ratio microchannels. The TTV is capable of dissipating 2600 W and measuring high-spatial resolution temperature data via 264 embedded platinum resistance temperature detectors (RTDs). The heat sink ’s performance with the refrigerant R1233zd(E) is assessed under aggressive heat loads intentionally selected to induce dryout. The resulting temperature signatures help deduce the areas over which dryout occurs, demonstrating the utility of the TTV in careful assessment of the thermal performance of two-phase heat sinks, especially under highly demanding thermal loads.
The increasing thermal demands of high-performance computing in AI datacenters, 5G radio access networks (RAN), and edge compute nodes necessitate the use of advanced cooling solutions beyond conventional air cooling. Multi-chip modules (MCMs), integrating CPUs, GPUs, and high-bandwidth memory (HBM) in 2.5D and 3D architectures, are now drawing power more than 1000 W. Assessing the efficiency of cooling solutions capable of dissipating such high heat loads requires developing a high-power thermal test vehicle (TTV) capable of mimicking the heat maps of the state-of-the-art and next generation CPUs and GPUs. This study demonstrates the high heat dissipation capacity of TTV while cooling it with a two-phase microchannel heatsink. At the core of the developed TTV is a multi-chip module consisting of six emulated logic dies (20 mm x 20 mm) and twelve emulated HBM stacks (10 mm x 10 mm) on a low-temperature co-fired ceramic (LTCC) interposer, capable of dissipating more than 2600 W with localized hotspots reaching 314 W/cm(2) (817 W/cm(2) over the heater resistor area). The module consists of 264 platinum resistive temperature detectors (RTDs) for high-resolution thermal monitoring and enables independent power control across 60 heater regions, facilitating diverse spatial heat maps. The TTV was integrated into a pumped refrigerant cooling system utilizing the low-global-warming-potential (GWP) refrigerant R1233zd(E) in a heat sink with embedded microchannels. Two-phase cooling successfully handled heat fluxes up to 314 W/cm(2) over a 46 mm(2) hotspot on each logic die while maintaining thermal resistances below 0.013 K/W, demonstrating its ability to prevent dryout and thermal runaway at extreme power densities. The experiments also highlighted the superior thermal performance of two-phase cooling over single-phase approaches, achieving up to a 2.8x reduction in mean junction-to-heatsink inlet saturation thermal resistance and reducing peak chip temperatures by up to 25 degrees C. This study demonstrates the utility of the TTV for emulating next-generation high-power computing platforms and the use of two-phase cooling as an advanced thermal management solution for future AI, 5G, and edge computing infrastructures.
Surface-normal electroabsorption modulators (SNEAMs) are devices with unique characteristics, such as small size, wide bandwidth and polarization insensitive behavior; however, due to the surface-normal configuration and since they operate in reflection, it is challenging to package fiber arrays to SNEAM arrays with many channels. Here, we present a novel approach to package fiber arrays to SNEAM arrays based on passive alignment. We realize an expanded-beam, multi-channel, optical coupling scheme by using a prism lens array and a lens array in between a SNEAM array chip and a standard single mode fiber array. We use this approach to assemble a SNEAM array engine with 8 channels without any active optical alignment. We show that our SNEAM array engine has coupling loss from the fiber inputs to the SNEAMs ranging from 1.2 dB for the best channel to 1.7 dB for the worst channel. Also, we validate our passive alignment packaging scheme by modulating the channels at 25 Gbit/s NRZ-OOK and show that the fiber array is pluggable into the SNEAM array engine. Finally, our approach is independent of the specific SNEAM design, and potentially can be extended to other surface-normal devices, such as lasers or photodiodes.
Oscillating heat pipes (OHPs) typically consist of a meandering or serpentine channel or tube that contains alternating liquid slugs and vapor plugs distributed along the length of the channel. Heat transport into and out of the channels results in fluid motion due to vapor bubble generation and growth (heat input) and vapor bubble shrinkage and collapse (heat removal), which, respectively, occur at the so-called evaporator and condenser sections of the OHP. OHPs are commonly used to transport heat laterally from a heat source to a heat sink. However, they can also be employed to effectively spread heat laterally within the plane of the OHP from hotspots on one side of the OHP to a heat sink on the opposite side; we refer to this configuration as a vertical heat spreading stack. The local vaporization and condensation processes inherent to OHPs allow them to mitigate unfavorable effects associated with high heat fluxes and localized hot spots where, for example, concentrated heat sources can result in large temperature spikes when relying on a purely conduction-based approach. This work focuses on evaluating the performance of OHPs in managing localized areas of high heat flux. Potential applications include electronics cooling and in-package thermal management. A vertical heat spreading test apparatus is developed to characterize the thermal performance of the aluminum OHPs with hotspots of sizes 4 mm x 4 mm and 4 mm x 10 mm and heat fluxes of up to 100 W/cm 2 . The effects of different working fluids and fluid fill ratios are studied. A performance map of the OHPs is presented that explores the dependence of thermal resistance on the hotspot size, location, heat flux, working fluid and fluid fill ratio.
We demonstrate, for the first time, fully-passive, optical packaging of an 8-channel array of surface-normal electroabsorption modulators to arrays of standard single-mode fibers. We validate our packaging approach through 25 Gbit/s NRZ-OOK modulation.
Short-reach communication systems use electro-optic transmission engines based on low-voltage, wide bandwidth modulators to achieve high capacity and low power operation. Here, we demonstrate, for the first time, integration of a 4-channel array of surface-normal electroabsorption modulators (SNEAMs) with a 4-channel array of electronic drivers with low output voltage (~1.25-1.45 Vpp). We use this 4-channel SNEAM-driver array electro-optic engine to demonstrate $4 \times 53$ Gbit/s non-return-to-zero on-off-keying transmission across up to 2 km of standard single mode fiber over a broad wavelength range (about 26 nm), the widest band of operation achieved with SNEAMs. Integration of a SNEAM array with a low output voltage driver array, a result that we achieved thanks to the use of SNEAMs that operate with low voltage swings, represents a fundamental milestone toward the realization of optical modules based on SNEAMs.
We demonstrate a WDM passive optical network that uses high-speed, polarization-independent surface normal modulators as upstream 25-Gb/s transmitters. A band of CW wavelengths, sourced at the optical line terminal, establishes the upstream channels. System performance in our single-fiber architecture is limited by Rayleigh backscattering, but is below the FEC threshold.
Short reach communication systems, such as datacenters and access networks, exhibit steep capacity growth and require opto-electronic technologies with small size, low complexity, and low power consumption. Here, we demonstrate large-scale arrays of reflective surface normal electroabsorption modulators (SNEAMs). With very small active volumes, SNEAMs enable ultra-wide electro-optic bandwidth (>>65 GHz). We show modulation at 25 Gbit/s with 1 V-pp drive voltage on packaged SNEAMs and ultra-high bit-rate modulation at 107 Gbit/s with bare chips. These modulators are polarization-independent and have very low total input/output coupling loss of 0.7 dB to single-mode-fibers. We package SNEAM arrays with arrayed waveguide gratings into wavelength division multiplexing transmitters. Due to their broad wavelength range of modulation, SNEAMs do not need power hungry wavelength tuning or locking systems. Future co-integration of SNEAM arrays with low-power electronic driver arrays will enable high-capacity, low-power electro-optic engines.
We measure performance of 44-Gb/s (22-GBd) PAM-4 short-reach direct detection links with a polarization-independent surface-normal electro-absorption modulator. Performance below the KP-4 FEC threshold is demonstrated in standard single-mode fiber links with no dispersion compensation for distances up to 18 km.
We report multi-level modulation in polarization-independent surface-normal electro-absorption modulators (SNEAMs). Four-level pulse amplitude modulation (PAM-4) at a line rate of 44 Gb/s is demonstrated on a fully packaged SNEAM with a 30 µm active area diameter and a 14 GHz electro-optic bandwidth. High-capacity PAM-4 transmission at 112 and 160 Gb/s is demonstrated on an unpackaged SNEAM chip, with a 15 µm active area diameter and ultrawide electro-optic bandwidth ( ≫ 65 G H z ). Fiber transmission is investigated for direct detection link lengths up to 23 km at 44 Gb/s and 2 km at 112 and 160 Gb/s, the highest multi-level modulation rates achieved on a SNEAM.
We demonstrate arrays of surface-normal electroabsorption modulators with ultrawide bandwidth (>>65 GHz), polarization insensitive response and ultralow total coupling loss to single-mode-fibers (0.7 dB). We show modulation up to 107 Gbit/s and packaging with arrayed-waveguide-gratings.
We present measurements of express optical interconnection between data centers using a MEMs photonic cross connect and a packet-optical transport network under the control of a programmable netOS for dynamic traffic steering and resiliency.
We developed a novel wavelength-stable burst-mode (BM) laser based on a multi-electrode distributed-feedback (DFB) laser. Our main use-case for the multi-electrode laser (MEL) is a time- and wavelength-division multiplexed passive optical network (TWDM-PON) (NG-PON2), but the wavelength-stable BM laser might also enable future coherent TDM-PON systems and high-rate PON systems using optical-filter-based chromatic dispersion compensation. Samples of the proposed laser are realized and characterized. The performance of the MEL-based BM transmitter is evaluated in several experiments in a directly modulated (at 10 Gbps) and externally modulated (at 25 Gbps) configuration. It is shown that the proposed laser is suitable for TWDM-PONs using a 100 GHz wavelength grid.
We report on a novel 10Gb/s low-cost multi-electrode DML employed as a very wavelength stable burst-mode source for upstream TWDM-PONs. 10X wavelength drift reduction is achieved compared to conventional DMLs enabling transmission on 100GHz grid.
Investigation of NG-PON2 upgrade to 25 Gb/s line-rate on a 100 GHz grid using a burst-mode transmitter based on a multi-electrode DFB. Compliance with NG-PON2 MSE requirements is shown.
This paper describes a compact and power-efficient video projector that can be readily integrated into mobile devices such as cell phones. Using state-of-the-art RGB lasers, micro-display panels, and advanced imaging techniques, this 6 cubic centimeter microprojector module can project video graphics array (VGA) resolution color video images that are similar in size and brightness to a notebook computer screen, while consuming only 1.5 watts of electrical power. The microprojector could be ubiquitously deployed as a preferred display device to connect, inform, and entertain the over 3.3 billion cell phone users worldwide. © 2009 Alcatel-Lucent.