Over the last 20 years, data rates for optical technology have climbed 1000x while power per bit has declined by 100x, a stunning trajectory that in many ways paved the way for the cloud, mobile Internet and streaming media.
AI represents the next inflection point in bandwidth demand. Servers powered by AI accelerators and GPUs have far greater bandwidth needs than typical cloud servers: seven high-end GPUs alone can max out a switch that ordinarily can handle 500 cloud two-processor servers. Just as important, demand for AI services, and higher-value AI services such as medical imaging or predictive maintenance, will further drive the need for more bandwidth. The AI market alone is expected to reach $407 billion by 2027.
O-band coherent or coherent lite—a technology that has been discussed for years at conferences but has yet to be deployed commercially in a meaningful way--will likely begin to percolate into the market over the next few years to help cloud service providers accommodate some of these challenges.
The Distance Dilemma
Optical DSPs, the chips inside the optical modules that power optical links within data centers, have been doubling in performance around every two to three years while the speed per lane has progressed at a similar rate. This year, we released Nova 2, the first 1.6T PAM4 optical DSP featuring 200 Gbps optical and electrical I/O lanes: a module powered by such a DSP can provide full bandwidth performance for links up to 10 kilometers.
With 400 Gbps/lane on the horizon, the ability to close links up to 10 kilometers becomes extremely challenging. And therein lies the rub: data centers will need to grow across multiple dimensions. Accommodating this increasing demand will not just likely mean more data centers. It will likely mean larger data centers requiring a greater number of longer internal connections, and even greater urgency for better performance and power per watt.
Companies are also looking at building virtual hyperscale data centers where modular units arrayed across a campus are linked together through high-speed fiber links. This approach mimics the performance of a supersized facility while curtailing the stress on water supplies and power that larger facilities can place on a region. Combine that with the anticipated growth in bandwidth and you can imagine the growing demand for high-bandwidth links spanning several kilometers.
A Change in Band
Coherent technologies have traditionally been used to connect geographically distant data centers, with connections spanning 100 kilometers and beyond. On the optical side, DWDM technology with tunable lasers operating in C-band is a must for long distance transmission in order to save the fiber cost and maintain performance. Operating at C-band, it is required for the coherent DSP to mitigate optical impairment such as chromatic dispersion (CD) and differential group delay (DGD). Intrinsically, the complexity of coherent DSP and the need to use tunable lasers and complex coherent optical receivers drive the cost higher, making it not viable for high volume deployment.
However, for shorter distance such as the typical 500m-10km interconnect for inside data center, building the optical network in O-band is more economical. Fixed wavelength lasers operating around 1310nm provide significant advantages on cost since dispersion is not a major concern. Since mitigation of CD and DGD are not needed in O-band, the complexity of the optical DSP can be simplified. The combination of using coherent technology with O-band optics could extend the reach, improve the bandwidth, and reduce the cost of the optical module. O-band will also likely be able to deliver the performance needed to replace standard coherent technology in campus environments to save energy.
Nothing stands still. Ten plus years ago, the industry standard was NRZ and PAM4 was an experimental idea. Now, PAM4 is pervasive and people have begun to build out the technology and use cases for O-band coherent. While we anticipate PAM4 technology will continue to evolve, the changing needs of data centers will create openings for O-band coherent. Although it’s difficult to predict how extensively O-band coherent will be used, we can predict that data centers will get larger and that bandwidth will continue to climb. Thus, even if it gets deployed in limited use cases at first, the potential opportunities will expand.
Stay tuned for more as we continue to explore all forward looking technologies to help the industry continue to increase the optical interconnect speed and reach, and drive the power and total cost ownership down.
This blog contains forward-looking statements within the meaning of the federal securities laws that involve risks and uncertainties. Forward-looking statements include, without limitation, any statement that may predict, forecast, indicate or imply future events or achievements. Actual events or results may differ materially from those contemplated in this blog. Forward-looking statements are only predictions and are subject to risks, uncertainties and assumptions that are difficult to predict, including those described in the “Risk Factors” section of our Annual Reports on Form 10-K, Quarterly Reports on Form 10-Q and other documents filed by us from time to time with the SEC. Forward-looking statements speak only as of the date they are made. Readers are cautioned not to put undue reliance on forward-looking statements, and no person assumes any obligation to update or revise any such forward-looking statements, whether as a result of new information, future events or otherwise.
Tags: Coherent DSP, Optical Module, PAM4 DSP
Copyright © 2024 Marvell, All rights reserved.