INDUSTRY NEWS
UCLA achieves optical computing boost with diffractive network advance
A recent publication in Advanced Photonics Nexus by a team led by Aydogan Ozcan, Chancellor’s Professor and the Volgenau Chair for Engineering Innovation at UCLA, have introduced a method to perform complex-valued linear operations with diffractive networks under spatially incoherent illumination. The paper explains that state-of-the-art neural networks depend on linear operations, such as matrix- vector multiplications and convolutions. While dedicated processors like GPUs and TPUs exist for these operations, they have limitations in terms of power consumption
and bandwidth. Optics is better suited for such operations because of its inherent parallelism, large bandwidth, and computation speed. Diffractive deep neu- ral networks (D2NN), also known as diffractive networks, constitute an emerging optical computing architecture. These task-spe- cific networks are construct- ed from spatially engineered thin surfaces and can passively perform computa- tional tasks at speed-of-light propagation through an ultrathin volume. The spatial features of these diffractive surfaces are learned through a one-time design process, and the optimized surfaces are then fabricated to create
the physical hardware of the diffractive optical network. Complex-valued Linear Transformations using Spa- tially Incoherent Diffractive Networks (a) The workflow of the spatially incoherent diffractive network model: a complex-valued ele- ment of the input vector is represented by a set of real, non-negative intensity val- ues (mosaicing). The result- ing input intensity pattern is fed into the incoherent diffractive network. At the output, a complex-valued vector element is synthe- sized from a predefined set of intensity pixels (demosa- icing). (b)Image encryption application. The letters ‘U’ and ‘C’ are encoded in the amplitude and phase of
a complex image, which is encrypted digitally and thereupon decrypted using the spatially incoherent diffractive network. The decrypted complex image matches the original image very well. Previous research by the same group showed that diffractive networks with sufficient degrees of freedom can perform arbitrary complex-valued linear transformations with spatially coherent light. However, under spatially incoherent light, these net- works can perform arbitrary linear transformations of input optical intensities if the matrix elements defining the transformation are real and non-negative.
Chinese transceiver vendors dominant in 2023 - report
LightCounting’s latest report, Vendors and Markets for Optics in China, finds that while Chinese manufacturers of optical components and modules reached a milestone in 2021, when revenues of the top 10 Chinese suppliers surpassed sales reported by their western competitors, the gap widened in 2022- 2023 in favour of the Chinese suppliers. Their initial success, says the report, was facilitated by strong demand for optics inside of China, but it was the sales of optics to the US-based Cloud companies that propelled Innolight and few other vendors to new sales records. However, customs duties imposed by the US govern-
ment on products made in China, including optical transceivers, forced many Chinese suppliers to estab- lish manufacturing sites in Thailand, Vietnam or other countries in East Asia. This required additional invest- ments, but it will pay off over many years to come. Lower labour cost in East Asia is another benefit, which may become even more impor- tant in the future, as many young people in China are no longer willing to take the factory jobs, which pro- pelled their parents out of poverty in the past. Steady investments in optical networking infra- structure in China created a US$2-US$3 billion domestic market for Chinese suppliers of optical transceivers. Light- Counting’s latest forecast
suggests that deployments of optical transceivers in China will account for 20-25% of the global de- ployments in 2024-2029, compared to 25-35% in 2018-2023. Very aggressive plans of US-based Cloud companies for deployments of 800G optics in AI Clusters is the main factor reducing China’s share in the next 2-3 years. The company expects that Chinese cloud com- panies and telecom service providers will be catching up with their western rivals in their spending on optics by 2027-2029. Another government priority which is impacting the decisions of Chinese cloud companies and tele- com service providers, says LightCounting, is the desire to allocate more business to
local manufacturers – not only of transceivers, but also of the laser and detector chips used in them. Despite significant progress made by Chinese chip suppliers, they are estimated to be two to three years behind their western competitors in the development of high speed (100G per lane) compo- nents. Lack of domestic sup- ply for 4x100G and 8x100G transceivers may limit their deployment in China for the next 3 years, but it is ex- pected volume deployments of these products in China to start in 2027. The report concludes however, that despite all the challenges, China will remain a signifi- cant market for suppliers of optical transceivers and a critical part of the industry’s ecosystem.
4
www.opticalconnectionsnews.com
INDUSTRY FOCUS 2024
Made with FlippingBook - professional solution for displaying marketing and sales documents online