- Show All
Shaping WTMF-2 in June 2018
The Story So Far
In June last year, more than 170 scientists, researchers, government representatives and high-tech industry experts from 17 countries converged on the Dutch city of s’Hertogenbosch to participate in the World Technology Mapping Forum (WTMF-1). The gathering was an important first step towards building the first international Integrated Photonics Roadmap, which looks beyond Moore’s Law to the world’s technology needs in 2030. As we kick off an important year for photonics in the Netherlands, Jonathan Marks caught up with Ton Backx, CEO of PhotonDelta, who saw the need for such a meeting of the minds, to reflect on the event and find out what the community can look forward to by June 2018 when WTMF-2 is held in Twente.
“We have capitalized on the extraordinary level of trust that was built up in Brabant”, Ton explains. “Since the meeting last June, we've been working out the details gleaned from the 6 parallel technical working groups that were formed. Based on that wealth of data and ideas, we've decided to redefine the structure a little bit, so that we can streamline the interactions, make the findings more coherent, and decide what is still missing.
We're doing this first report together with Berenschot, who are well-known technical strategy consultants in this part of the Netherlands. Together, we've been learning from the experiences of our colleagues at AIM Photonics Academy who set up the IPSR International. At the same time, we’ve been making sure that we're setting up a collaborative process that is manageable and yet has short communications lines.”
“Berenschot has been in touch with each of the six working groups and has been filling in the results according to the structure we agreed upon. They are nearly finished with what is a very complex task. I hope that by the middle of January 2018 we will be able to release the first draft of the Global Roadmap. It will then be opened for on-line discussion amongst those who attended the first World Technology Mapping Forum and any other parties who believe they can contribute. Of course, I don't expect it to be complete because our discussions in June this year were simply a starting point. But it a good start for the next WTMF-2 which is coming up June 20-22nd 2018 in Enschede, the Netherlands”
Starting an important process
“To keep discussions going, we also have fortnightly collaborative meetings with our colleagues at AIM Photonics and the IPSR. They have their established roadmapping procedure centered around silicon-based photonics and we actively participated in their fall AIM Photonics meeting in October 2017 to bring in the Indium Phosphide expertise”.
“I believe we still need to fine-tune the structure of the conversation to ensure that not-only research goals are covered well from a technology perspective, but also that everyone has a clear understanding of industry needs and expectations. In that way, the roadmap will show what kind of functionality can be expected at what point in time to produce new products. We also need to forecast what sort of cost levels can be expected, so we can check whether these market propositions are feasible and look for tipping points. Our roadmap needs to support both these lines of thinking”.
“Integrated Photonics has developed using various materials for the wafers. In the Netherlands the expertise is centered around indium phosphide and silicon nitride. Other countries like Belgium, France, the USA, and Taiwan have a different focus on silicon photonics, which is the material used in the existing micro-electronics industry. Several different hybrid solutions are being developed now for next generation applications, which will mix different materials. That has several advantages, but will also raise all kinds of new connection issues between the two”.
Looking five years ahead, we foresee the need for radically different designs by 2023. That's because there is mismatch between the energy consumption of large datacentres and the local energy supply. We will need to build distributed datacentres and that means we will need to design much longer distance interconnects between them.”
Investors and industry interest
Many parties are closely watching the outcome of the deliberations from the technology forum. Large private and public investments will need to be made to scale up the manufacturing volume of next generation chips. Investors are naturally keen to find, select and back the winning technologies that have a proven product-market fit.
The photonics manufacturing industry wants to understand the technology choices that have been made to bring next generation products to market at a price that customers will pay for. These companies don't want to be confronted with a range of technological solutions for a product that they want to launch into the market next year.
Why silicon and indium phosphide need each other
“If we look at next generation devices that will come on the market this year and 2019, you will see several competing solutions from the separate silicon and indium phosphide worlds. Globally, a lot of work is being focused now on the silicon photonics ecosystem. But one of the drawbacks of silicon is that you can only produce passive circuits. When the circuit complexity is relatively low, you can make use of external discrete light sources or simple light sources integrated on chip. There are many manufacturers who can supply the discrete lasers. In the short-term there is no problem. But let’s look ahead 5 years.”
A world of 1600 Gb/s
“By 2023 the global market will be asking for transceivers that operate at a speed of 1.6 Terabit per second. If we take that as an end-goal, then compare it with the situation today where 0.1 Terabit per second is common. Today, communication links between datacentres all use optical links with single mode fibres. But the processing of the data in the data centres is done electronically.”
“The problem is that when you go from 0.1 to 1.6 Tb/sec, the light part can cope relatively easily, but the speed of the electronics becomes a severe bottleneck. With current technology, you can find commercial electronics modules operating at speed of 10 Gb/second. To get a transceiver operating at 100 Gb/sec, you build ten 10 Gb/sec electronic channels and operate them in parallel. The optical side already works at 100 Gb/sec. There is, of course, a conversion process to split the optical signal over the 10 parallel electronics channels.”
“Suppose we go to a transceiver operating at 400 Gb/sec, then this will require 40 parallel channels on the electronics side if the same electronics is applied. That is a lot more complex to achieve.”
Solutions for the bottleneck
Electronics is still important in the coming years. But it is also a bottleneck and it affects the way systems are going to be engineered.
“To come up with 400 Gb/sec transceivers for launch in 2019, some manufacturers apply electronics operating at 25 or even 50 Gb/sec. That would require only 8 parallel channels instead of 40. But this doesn't solve the problem by the time we've moved to a world of 1600 Gb/sec by 2023.”
“Having 160 or even 24 electronic channels operating in parallel raises the question where the conversion will be done. If the splitting, processing channel conversion is done on the optical side it can be done very efficiently, but additional signal processing complexity is required at the optical side. CMOS technology-based electronics will become significantly more complex when it has to operate at bit rates beyond 25 Gb/s”
“Many parties working on the 400 Gb/sec solutions with silicon-based PICs have concluded that for the moment, the chips will do limited optical processing. They will make use of a single or a few external laser sources to provide light to the silicon transceiver. The systems operating in that way will not have very long-distance communication capabilities. The distance they can bridge is not much more than 2-3 kilometres. While that may sound a lot, next generation distributed data centres require high speed transceivers that can bridge distances of 50-100 kilometres. Indium phosphide transceivers have already bridged a distance of 50 kilometres and beyond. So it may be expected that next generation transceivers applied for short distance may predominantly be based on silicon based photonics fed by an external laser while indium phosphide based high speed transceivers will be used for longer distances and short distance.”
Exponential energy challenges mean disruptive solutions
“Looking five years ahead, we foresee the need for radically different designs by 2023. That's because there is mismatch between the energy consumption of large datacentres and the local energy supply. We will need to build distributed datacentres and that means we will need to design much longer distance interconnects between them.”
“Let me share some numbers that came out of the 2017 WTMF technical workshop discussions. We expect that the 1.6 Tb/sec transceivers we will be using by then will need to bridge distances of 50-100 kilometres. Datacentres used for cloud functionality are being built by enterprises like Amazon, Google, Apple, Facebook, and Alibaba. They approximately have to double the capacity every year.”
Google Eemshaven in 2016
“You may know that Google has built a €600 million datacentre in Eemshaven in the North of the Netherlands. They report they have laid 16,000 kilometres in cabling. Now, although the energy being supplied to the new facility is currently 100% from renewables, the power consumption is around 250 Megawatts. The power station in the area only has a capacity of 1.2 GW available. But at the current rate of growth using conventional technology, the demand will be around 500 Megawatts by the end of 2018, and close to 1.0 GW by the end of 2019. That puts a very difficult demand on local power resources. Doubling the generation capacity of that power station is going to take seven years, so that is not going to solve the issue. So, you will have to split up that datacentre capacity, so it can tap into other energy resources from a region nearby.”
The impact of photonics
“The good news is that the introduction of photonics into datacentres will impact their power consumption quite considerably. We estimate that instead of doubling each year, power consumption will go down to a factor of around 1.3 - 1.7. But it will still require more datacentres operating in parallel to satisfy the exponential global demand for bandwidth and processing power.”
“There is another trend we need to consider. As the complexity of these next generation chips increases, light losses in passive light processing need to be compensated for using light amplification and additional light sources. That puts limits on the chip's functionality,if only passive circuits can be used, because at a certain moment, all the light is gone. Active laser sources on the chip will be needed to regenerate and amplify the light. We believe that indium phosphide is going to be needed in all the scenarios because when you need hundreds of laser sources or light amplifiers, you are not going to be able to use external discrete lasers anymore. By 2021/2022 we need to have a clear structured plan of how industry is going to do this, so roll out of the second chip generation is ready by 2023.”
“This realization that the current technology cannot keep pace that motivated me to start travelling round the world to urge others in photonics to collaborate and solve this pressing problem. I'm pleased to say that by openly collaborating we are getting much closer to a solution. Our colleagues at AIM Photonics Academy in Boston are doing some excellent work. And in Europe there are around a dozen hotspots that have put photonics near the top of their high-tech agenda. You will hear more about the European Photonics Alliance next year. PhotonDelta and AIM Photonics Academy are already convinced that working together is going to be essential to solve the packaging issues and to build very complex chip structures with active circuits on board. Joint research and knowledge exchanges are being planned for the middle of 2018 because there is no time to waste.”
Moore's Law is no longer fast enough for photonics
PhotonDelta recently announced the launch of a national Photonic Integration Technology Centre (PITC) as an important part of a new Dutch National Photonics Agenda.
“This is the result of three photonics “centres of excellence” working together, in the areas around Enschede, Eindhoven and Nijmegen. Each region has complementary micro-electronics knowledge and PIC expertise. And by global standards we're physically very close to each other. The PITC will offer its services through new alliances currently being established all over the world. It's a global organization with a Dutch accent, encouraging Open Innovation development.”
“Manufacturers, integrators and end users all have the same goals. They want reliable, tested, stable systems at the lowest possible cost. PhotonDelta's PITC will play a leading role helping companies accelerate in these very important packaging and reliability-engineering phases to retain and grow volume manufacture of photonics devices in Europe. The PITC will support the chip fabs like SMART photonics, Heinrich Hertz and LioniX International to ensure that their second-generation chip production technology can scale at exactly the moment there is the right product-market fit. The PITC will also do the further development in materials processing and systems development to support innovative enterprises like Technobis and EFFECT Photonics.”
PhotonDelta Cooperative established
“We have also established a separate PhotonDelta Cooperative as a new independent trusted, legal entity. Research organizations and industry are being encouraged to join to get early access to fundamental and applied research into next generations chips, circuits and systems and the associated IP. This is essential to ensure that we maintain and grow the high-pace of research that is needed now, but also that we keep listening to understand what the market needs tomorrow. The party that pays gets ownership of the IP, but the IP generated in that Open Innovation structure must be made available license-free for further research. If we get bogged down in endless IP ownership discussions pace of development will be too low whereas speed of development is critical in supporting the market. ”
“We have learned a lot from from our Canadian and Irish colleagues about the need to encourage cross-fertilization between research, development and industry. We need to encourage people to build a career across these sectors, in the same way this was initiated in microelectronics in the 1970's. So, you might start in fundamental research. Once some of the researchers have validated a technology they can move on to application development within an organization like PITC. Later they would go to work within industry to ensure the faster execution of that technology. With product life cycles so short these days, we think this is the best way to speed up knowledge transfer. If we keep a simple and transparent structure, then it is much easier to make sure it works!"