Dec. 30, 2024
Photonic Computing
Exponential growth in AI models has shifted focus from the "System-On-Chip (SoC)" concept to "System-Of-Chips." For both AI training and inferencing, it is imperative to interconnect multiple GPUs to provision the required computing & memory resources. AI data centers have typically implemented GPU-to-GPU communication using scale-out or backend-networks using Ethernet (RoCE) or InfiniBand. However scale-out networks which typically connect O(10000) GPUs based on 400/800Gbps networks do not provide enough bandwidth at low latency for efficient GPU-GPU communication. Power consumption of scale-out networks is also an important consideration as data center power requirement is a major factor in AI growth. A new class of networking is emerging called “scale-up” network which connects O(10s-100s) GPUs with a very high bandwidth, low latency network. The requirements of these scale-up networks, their design, implementation, and performance differences compared to traditional backend/scale-out networks, and why silicon photonics technology is best suited to implement these scale-out networks, are the focus of this article from optical compute solutions developer Celestial AI.
Key Technologies: Silicon photonics, optical interconnects, AI, optical computing, pluggable transceivers, active optical cable, co-packaged optics, linear-packaged optics, optical I/O
Laser Communications
Laser communication terminals (LCTs) deliver ultra-fast, secure data transmission, transforming satellite constellations, science missions, and defense operations. The technology has matured, and it is now finding use as an enabler to applications, moving beyond experimental trialing and proofs of concept. This article examines the design and functionality of LCTs, and looks at how their development and rollout represents a direct support for a network of lasers in space. The need to cultivate sustainable, scalable LCT manufacturing protocols will be imperative to maintaining the recent wave of momentum in laser-based communications, and the article shines light on some of the key players, from all reaches of the supply chain, already contributing to his need.
Key Technologies: Laser communications, laser communication terminals, space communication, data communication, piezoelectric elements, space optics, photodetectors (avalanche and single-photon), telescope optics, optical telecom
Photonics and the Data Center
The race to 1.6 Terabits/second and 3.2Tbps are critical for the next generation of computing solutions for the data center industry. Hyperscalers, which are increasingly driven by demand from AI hardware and software developers, require far more bandwidth than ever to satiate the global desire for more technological advancements in sectors such as medtech, wearables, robotics, automotives, and more. Large language models (LLMs) that are at the root of AI development can only get smarter by having more data and the ability to compute that data faster. So, the race to 1.6T is afoot and the crucial means to getting there has been revealed. The optical interposer has become the rising star in chip-scale architecture because the capability of reaching the next level of data speeds without exploding costs or power consumption. In fact, because the best optical interposers on the market do not require wire bonds or active alignments, they can drastically reduce costs and produce coveted power savings. Dependent on the efficient performance of lasers, optical interposers are likely to power datacenters into the AI era and beyond.
Key Technologies: Optical interposers, optical interconnects, optical datacom, integrated lasers, PICs, optical computing, optical modulators, integrated waveguides, electro-optics
Optical Networks
By 2030, the data demands of individual consumers will push current network infrastructure to its limit; the global average data usage per mobile user stood at approximately 8GB per month in 2022, but it is projected to surge to over 75GB per month by 2030. A number of drivers, from the adoption of data-intensive applications such as high-resolution video streaming, cloud gaming, and augmented/virtual reality, to the boom in AI are behind this forecast. Masahisa Kawashima, NTT IOWN technology director, identifies the latency and power limitations of existing electronic networks and introduces data center/power usage concerns related to necessary levels of energy for processing, as well as cooling and transmission challenges. The solution, all-photonic networks (APNs) achieve ultra-high capacity with data processing of 125x greater than networks today (by volume); ultra-low latency offering near-instant transmissions with end-to-end latency reduced by over 200x; and ultra-low power consumption with a goal of 100x more efficiency than current transmissions today, signaling a reduction in carbon emissions by 45%. Kawashima further discusses APN implentation challenges and projections, as well as scaling.
Key Technologies: optical/photonic networks, optical computing, optical data and telecom, cloud computing, edge computing, optical transceivers
EPIC Insights: Quantum Computing
Quantum computing, while still maturing, is receiving significant funding and making strides towards becoming a practical technology. The industry is transitioning from concept to reality, with end-users in cryptography, finance, medical, military, and industrial applications. The quantum technology market is projected to grow from $761 million in 2022 to $2.13 billion by 2030, with quantum computing being the fastest-growing sector. Various photonics-based architectures for quantum computers are under development. Technological advancements are anticipated in superconductor, ion-trap, spin-qubit, and photonics quantum computing. In fact, the photonics approach receives the most funding. Here, we investigate several recent advancements in laser systems, fiber alignment technology, and photonic processors for quantum computing in Europe.
Key Technologies: Quantum photonic computing, quantum optics (spatial light modulators, photondetectors and photodiodes, waveguides, beamspliters, optical switches), quantum photonics processing, lasers for quantum, optical fiber arrays, PICs
Download Media Planner