This past October, Ciena announced its new coherent DSP, dubbed Wavelogic Ai. Ciena used the announcement to convey its coherent 100G+ roadmap and support for 400G. Ciena also hinted at important changes to its 100G+ business model as Wavelogic Ai is introduced to customer networks. These changes also cascade into the 400G component supply chain.
Wavelogic Ai doubles down on Ciena’s strategy of vertical integration of the DSP and optics, and extends it further into the control plane. The company continues to eschew the use of outside suppliers like Acacia, NTT Electronics, or Clariphy. We believe Ciena’s vertical integration is the company’s strategy for successfully competing with commodity hardware and dis-aggregated networks.
Wavelogic Ai follows on the heels of the Wavelogic 3 announced in 2012 and incremental iterations (Extreme and Nano) announced thereafter in 2015.
It is expected that products based on Wavelogic Ai will start to appear in 2017, with a technology release around Q2. We have not yet spoken to customers about Wavelogic Ai, but we have seen some of the core technology firsthand in the Ciena labs.
Ciena makes the proposition that there is just too much margin in some coherent networks, and opportunity can be realized when technology that can sense and adapt to optical link characteristics squeezes out more capacity. Wavelogic Ai is designed to optimize this capacity * distance * spectrum product over all conditions, rather than forcing the use of hardware tailored for one reach. What is notable about the Ai is that it will also link with the optical control plane, and be implemented in a way that helps solves the flex-coherent pricing problem.
Flex-Coherent Priced to Market
The promise of flexible coherent interfaces has always been one in which a single piece of hardware could be used for any reach. But this promise has not yet been realized. The cost of optics differs depending on the required performance, and no one wants to pay for a 200G capable link that runs at 100G. Customers – particularly in DCI applications at Cloud and Colo vendors – don’t want flexibility if it can only be attained via a premium price.
Ciena’s plan is to pair Wavelogic Ai with universal optics that work the full range of speeds and baud rates, and then dynamically price the solution according to the speed used. Hardware designed for high performance, for example, could operate at slow speeds but also at a lower cost to the network operator.
Infinera pioneered this approach with its Instant Bandwidth product, where latent capacity can be turned on via a software license. This allowed customers to deploy 500G of capacity without paying for it up front. This capability becomes more useful with the company’s newest GEN4 PIC technology, which operates at multiple speeds.
While the exact implementation details of Ciena’s approach have not been announced, we expect Wavelogic Ai will create demand for 400G components faster than what we previously expected, since customers will be able to deploy 400G capable hardware without the upfront cost of paying for cutting edge hardware.
The primary supply chain beneficiaries are companies like Inphi, Lumentum, and Oclaro who invested early in 400G component technology. And rather than demand for 400G components being tied to only 400G speeds, with Wavelogic Ai 400G capable components are needed regardless of the line card operating speed. This steepens the initial demand curve for 400G components.
There are a number of interesting technology data points and specs with Wavelogic Ai, as follows:
- A new 28nm silicon on insulator fab process that cuts power in half. Power was never disclosed for Wavelogic 3 series components; but despite the use of “older” 28nm silicon, power performance is improved.
- Max symbol speed is 56 Gbaud – this is really fast. The Nokia PSE2 solution announced this year is 45 Gbaud. The Ciena solution can adjust the baud rate down to fit channels within 50Ghz channel spacing if needed, or run it up if the network can handle a non-ITU grid. The company claims Wavelogic AI’s higher baud rate nearly doubles the capacity of a link over a given distance–obviously mileage will vary depending on the characteristics of the given link.
- Ciena wants to de-emphasize the atomic nature of individual speeds and modulations and remove the need for the operator to provision or even be aware of them. This is partly because the company would rather not broadcast the specifics of how it achieves its performance. Ciena did indicate that the device uses multi-dimensional coding techniques and non-traditional constellation types.
- The fast symbol rate should result in superior 400G performance, and the company claims up to 300km of reach. We imagine that a flexible baud/modulation approach at 100km uses a tighter wavelength spacing than at 200km or 300km. Likewise, Ciena claims 200G at 3000km, essentially doubling capacity at distances that would use QPSK. Subsea is also in the Wavelogic Ai toolbox with 100G at extreme trans-pacific distances. Fast and flexible baud rates open up a lot of possibilities.
- A central tenet of Wavelogic Ai is to build on Ciena’s history of having a great deal of visibility into link performance via electronics, and using this information at the system and network level to self optimize. Wavelogic Ai enabled links will examine TX/RX optical power, linear and non-linear noise, current chromatic dispersion, and pre and post FEC error rates. It will then adjust a multitude of variables such as baud rate and modulation, and others that we speculate could be channel spacing, Nyquist transmit pulse shaping, and TX power to extract the best transmission rate. Ciena has always been obsessed with tuning the optics (Such as type-III ACOs), and this chip fits the bill.
DSP Paired to Control Plane
This obsessive pursuit of link data provides the core technology for an interesting step beyond current transponder-centric coherent approaches. We expect Wavelogic Ai will pair with Liquid Spectrum (Whitepaper link), which is Ciena’s up and coming optical control plane solution. Liquid spectrum will have access to coherent parameters from Wavelogic Ai and could theoretically use that information to optimize at the network level.
One can just imagine Liquid Spectrum pulling information from the Wavelogic Ai to adjust the filter profile of a mid-span ROADM hundreds of kilometers away in order to improve optical channel isolation. Or, it can provide more precise measurement of performance in the network to allow more optimal capacity and route planning. Many theoretical possibilities are inherent with this vertical integration, but making them work is no slight undertaking. The key observation – without vertical integration they are not possible at all.
Fighting the White Box Threat
For a long time, network hardware has been vertically integrated with a single vendor providing an end-to-end solution. However, network operators will soon have the option of cobbling together optical networks from multiple vendors.
This is the dis-aggregated network hardware model, in which equipment from multiple vendors, pluggable coherent modules, ROADMs and other common equipment are all provided by the lowest bidder or selected on the basis of superior performance. Operators will still require the software to link and control all of these disparate vendors, and companies like Google already do this while others such as AT&T are trying to get there.
Through Wavelogic Ai and potential optical control plane integration, Ciena can extract the maximum value out of the vertically integrated vendor model. It must do this, because funding the design of the DSP, the hardware, and the software supporting such a complex optical control plane is a big investment for the company. If Ciena can ship enough volume to offset its significant fixed cost, it can beat the commodity hardware model in most end customer markets.
If someone purchases exclusively from Ciena, they will be buying a network that outperforms commodity equivalents. That doesn’t mean Ciena cannot or will not sell into dis-aggregated networks, but Ciena’s proposition is that, for certain customers, end-to-end ownership and control yields better performing price/performance than dis-aggregated optical networks can provide.
This is where Ciena’s marketing message of “too much margin” in the network is rooted.
The link of high volume 400G components with 56 Gbaud 400G capable DSPs and an optical control plane is an impressive end-to-end solution that can only be achieved by vertically integrated equipment companies. We expect Ciena’s approach to be successful in extracting more capacity from complex networks compared to dis aggregated hardware requiring more margin for reliable performance.
We at Cignal AI admit, this article sounds quite pro-Ciena. Our enthusiasm is bolstered by Ciena’s track record of being first to deliver production coherent technology and commitment to investing in software. Companies with deep technical skills stumble all the time – Ciena has its share of mistakes too. We’ll see where Ciena ends up with 400G.