Thanks for visiting Fibre Systems.

You're trying to access an editorial feature that is only available to logged in, registered users of Fibre Systems. Registering is completely free, so why not sign up with us?

By registering, as well as being able to browse all content on the site without further interruption, you'll also have the option to receive our magazine (multiple times a year) and our email newsletters.

LightCounting: machine learning will boost optical interconnect markets

Share this on social media:

The market for active optical cables (AOC) and embedded optical modules (EOM) is set to experience unprecedented growth, according to the latest report from specialist market research and consulting company, LightCounting.

Now in its tenth edition, the study has forecast a likely rise of 33 per cent in unit shipments by 2021, in comparison with the company’s research a year ago. In spite of this, however, the revenue forecast for 2021 has only risen very slightly as prices erode.

According to the firm, new opportunities are arising as the result of an important trend in high-performance computing (HPC) and hyperscale data centres - machine learning. Deep neural networks are needed for, amongst other things, better image recognition. There is so much additional content to categorize; it cannot happen with current technology.

Machine learning is about tightly coupled racks of hardware. ‘Between these racks, the coupling is loose - at least for now,’ explained Dale Murray, principal analyst at LightCounting. ‘Machine learning places heavy demands on interconnects to the processors and to memory, creating a push for higher performance and more advanced technology over cost alone. It will change the physical architectures from stacked servers with leaf and spine switches to entirely new fabrics.’

One good example is a Gen-Z, a memory-centric switched fabric architecture defined by a consortium. Gen-Z is all about making new ‘storage class memory’ available to each CPU in the system. Shared DRAM and MRAM allow each processor to see a large pool of memory as a continuation of its own local memory. Gen-Z today runs at 25Gtransfers/s. It will move to 56G and 112G in the future. Optical interconnects will come into play at 56G, says LightCounting.

For nearly 20 years, the EOM market has served high-speed, intra-system proprietary links inside custom supercomputers, telecom equipment and the largest core routers. The HPC market – which generates the greatest demand for embedded optical modules – is sensitive to large orders and is poised for new growth, according to the market research firm. The fastest machines in the Top 500 list of supercomputers will be turning over with exciting new machines on the horizon.

The firm now projects the 2021 EOM market to be around 6 per cent smaller in unit shipments but 11 per cent greater in revenue than forecast a year previously. It says annual growth rates will increase strongly by 2019.

This AOC/EOM report and forecast is an update of the December 2016 report of the same name. The report explores recent and future applications in HPC, cloud data centers, core routers, telecom equipment and military/aerospace markets. Historical shipments are explained for 20 product categories. A detailed Excel spreadsheet accompanies the report, providing a five-year forecast of shipments, prices and revenues broken down by product types, speeds and by application.

Company: 

Related news

Recent News

01 June 2020

A team of researchers from RMIT, Monash and Swinburne universities in Australia, has achieved a data speed of 44.2Tb/s from a single light source.

27 May 2020

Broadband Forum has made significant progress on bringing an improved broadband user experience closer, with two new specifications nearing completion.

26 May 2020

Openreach engineer Scott McPartlin pitched in to help remote, vulnerable residents on the island of Coll, stay connected during lockdown.

19 May 2020

Microsoft has signed a definitive agreement to acquire virtualized network software provider, Metaswitch Networks.