Optisk computing kan (muligvis) løse AI-applikationers powerkrav
Optisk computing forventes at blive en realitet i en ikke fjern fremtid takket være store fremskridt inden for siliciumfotonik og kvanteoptik. Analysefirmaet Yole belyser udviklingen på området i ny rapport (in english).
Optical computing is still in its early stages. While some major companies have shifted their focus from optical computing to optical I/O, startups in optical computing continue to emerge, exploring various approaches. Optical processors are primarily targeted at AI acceleration and inference tasks but also high-end niche applications, while optical quantum computers utilizing qubits and other quantum effects are aimed at applications like simulation, optimization, and AI/ML .
Eric Mounier, Ph.D., Chief Analyst, Photonics & Sensing at Yole Group explains:
- The first optical processor shipments will appear around 2027/28, initially for custom systems implementing parts of the technology, with revenue largely coming from NRE services. By 2028, general-purpose optical processors would enter the market, with early adopters and OEMs gradually integrating them from 2029 onward. By 2034, at Yole Group, we expect optical processor shipments to reach nearly 1 million units, representing a multi-billion-dollar market.
Additionally, photonic-based quantum computers are expected to see substantial growth starting in 2030, led by companies like Quandela, QUIX, and Pasqal, with the market projected to be worth hundreds of millions USD by 2034.
In this context, Yole Group launches its new photonic report, Optical Computing 2024. This report provides a comprehensive overview of the current status and future prospects of optical computing, focusing on both analog/digital optical systems and quantum optics approaches.
This new analysis explores how these technologies can address the growing computational needs of AI, ML, and other usages by overcoming the limitations of traditional electronic systems. This report is part of the significant photonic collection of Yole Group’s analyses, including Silicon Photonics 2023 (2024 edition coming soon), Co-Packaged Optics for Datacenters and more.
Optical computing is not a new concept, and there are various ways to implement optical gates, with photonic ICs and quantum optics currently the most promising. However, despite advancements, practical optical logic gates still face considerable challenges. To compete with electronic gates, they must achieve crucial criteria like cascadability, scalability, and recovery from optical losses.
While much of the current research is focused on single gates or simple circuits, large-scale optical computers are still in the early stages of development. Silicon photonics is seen as a key enabler for optical computing due to its scalability, though integration has been a persistent challenge in photonics.
Advances in integrated optics, using materials like SOI , SiN , TFLN , graphene, BTO , and polymers, could lead to the development of practical optical processors based on photonic ICs. These improvements will also benefit quantum optics, facilitating the creation of quantum optical computers with more qubits in a compact form factor.
There are multiple approaches to creating an optical processor, either analog or digital, using different optical media such as photonic ICs, FSO , or fiber optics. In the realm of optical quantum computers using qubits, three main approaches are being explored: one using photon qubits and the others employing photonics to control non-photonic qubits, such as trapped ions and cold atoms.
Some companies are developing optical quantum computers that do not rely on qubits, instead using optical quantum effects and nonlinearity. Novel materials like metasurfaces and SiC are also being investigated for optical processors, though they are still in the early stages of research.
