Optics accelerates footprint within data centers, advanced vision applications as AI surges

By Todd Haugen / Enablence Technologies

Recent advances in AI are challenging infrastructure in terms of capacity, throughput, power consumption and heat generation.  Historically, the chip industry looked to smaller architectures to address these problems, commonly known as Moore’s Law.   

The industry has now reached the operational limits of advancements in electronics with the advent of 4nm technologies being rolled out by leading manufacturers.  The industry is increasingly looking to optics for continued advancements in throughput and latency. 

PLC, SiN, SiPH optics technology offer alternative energy efficient paths

The promise of optics chips for computation, among other functions, is emerging as a significant technological advancement that overcomes the limitations of traditional electronics.  Optics are superior in speed, energy consumption, and bandwidth while generating minimal heat. 

Optics processors operate efficiently at room temperature, minimizing the need for the specialized cooling systems found in data centers employing traditional CPUs and GPUs.  Optics technology is based on three platforms, Planar Lightwave Circuits (PLC), Silicon Nitride (SiN), and Silicon Photonics (SiPH).  

Nvidia, and others are now looking to a world where optics removes the bottlenecks in infrastructure that AI is laying bare for all to see.  For example, optical transceivers once limited to communications functions within the datacentre are now expanding to support interconnect and compute functions that support AI and the development of Large Language Models (LLMs). 

As optics expands its footprint throughout the datacentre, companies like Enablence, Lumentum, Infinera and others will continue to realize major new growth opportunities.  In addition, the emergence of AI is also accelerating the deployment of advanced visons systems technologies like LIDAR with the potential to speed advancements in medical, aerospace, defence, and automotive sectors.  Today, though AI is transforming datacom and growing the optical industry in several ways.

• First, Overall network capacity – training large language models (LLMs), or other Generative AI models requires massive amounts of compute to crawl available data sources to train models and increased bandwidth to make the resulting models available to users.  This increased network demand is increasing demand for networking optics.

• Second, High bandwidth, Low latency Intra-rack connectivity – LLM training is performed by data centers with racks upon racks of GPU pods (AI compute) that need massive amounts of high bandwidth, low latency interconnectivity.  This has resulted in the rapid emergence of co-packaged optics (CPOs) and NxN routers to address this growing requirement. The rate of GPU deployment to provide AI compute is accelerating at a phenomenal rate evidenced by Nvidia’s increased production volumes which is increases interconnect demands commensurately. According to Statista Research, the GPU market will grow 31 percent between now and 2029 exceeding $269 billion. According to Global Market Insights, the Optical Transceivers valued at $13 billion will surpass $25 billion by 2029 growing at CAGR of 13 percent through the period. Interconnect demand will see a doubling of the market from $11 billion to $26B billion by 2030 says Mission Critical Intelligence.  
  The optics market overall will grow from $287 billion in 2023 to more than $628 billion in 2032, especially as AI intensifies and Moore’s Law limitations become reality, the industry will increasingly turn to Optics as one of a handful of remedies to meet the rapid demand for AI.

• High throughput, low power AI compute inside the chassis – several companies, such as Lightmatter.ai with its Envise product line, are bringing optics inside the chassis, significantly increasing in throughput while dramatically decreasing power consumption and heat generation.

The demand for optics generated by AI in these three areas is reshaping how optics companies invest in R&D. This will drive unprecedented growth over the next ten years.  Within the datacentre, optics, with its inherent power efficiency, heat, and capacity benefits, will expand its footprint markedly, especially as AI demand continues to surge. Optics device suppliers like Infinera, Lumentum, Enablence and Viavi are already working with customers in these three areas to accelerate and support this transformation. For example, Enablence’s exceptionally high-power handling capacity, exceptionally low loss, and efficient use of space delivers interposers and flat performance NxN routers previously not thought possible in the industry.  Additionally, Enablence is today helping AI partners bring solutions to the market in record time through 6–8-week cycle times, requirements to tested wafers especially. Beyond the surge in demand for optics from Enablence’s traditional datacom and telecom markets, accelerated interest and demand for advanced vision products is also being spurred on by AI and LiDAR technology developments.

Optics has now become a core component of advanced vision which is spread across medical manufacturing, aerospace and defence, automotive, drones, robotics markets and much more. These sectors all use major components of advanced version, which can be defined as anytime you shoot a laser into free space. We then collect the photons that bounce off the environment and process them to understand the surrounding environment. That sums up advanced vision, and for example, within advanced vision you have you've got short range applications like medical. These applications can see millimetres or centimetres into the skin to understand things like whether a biopsy is required. Time to disease diagnoses should accelerate.  Beyond accelerated disease diagnoses, self-driving cars, drones, advanced robotics, planes are all within reach.

Today, manufacturers can turn to a myriad of hybrid optics solutions that maximize the capacity and efficiency of PLC technology and combine it with the miniaturization and active integration of SiN and SiPH platforms. These hybrid solutions enable optics to address communications, interconnect and the complex computing tasks demanded by a surging AI marketplace and in the process help to usher in new wave of technological innovation.