Silicon photonics nearly with us, says Intel

Transporting these levels of data around through traditional on board interconnection methods has limits in terms of bandwidth, but also in terms of power consumption and physical size – even weight. Intel has been running its Silicon Photonics Operation in a bid to find solutions to these problems by using photons, rather than electrons. Business development manager for Intel's Silicon Photonics Operation, Jeff Demain, said the driver has always to bring the bandwidth that lasers can provide into the modern computing and communications environments that badly need it. "But it has been the classic 'chicken and egg' scenario – never enough volume to reduce costs, nor cheap enough to become high volume." Silicon photonics, by combining laser performance with low cost silicon manufacturing techniques, has therefore been the goal towards which Intel has been working. Nearly a decade ago, the breakthroughs came with the first continuous wave silicon Raman laser and the hybrid silicon laser, silicon modulators of rapidly increasing speed (40Gbit/s in 2007), 10Gbit/s PIN photodetectors and the 340GHz gain avalanche photodetector in 2008. Given the amount of data needing to be moved, Intel saw the first target as data centres and, last year, demonstrated its Rack Scale Architecture (RSA), with silicon photonics working in conjunction with the newly defined MXC connector. The functional rack featured Atom C2000 and Xeon CPUs. Optical PCIe was also demonstrated last year and, in April this year, Huawei demonstrated its router board featuring Intel silicon photonics. And, while the full launch of the technology is unlikely before the autumn at the earliest (look out for news at the San Francisco Intel Developer Forum in September), Demain believes we are on the verge of a breakthrough. "What has made this possible is being able to manufacture billions of transistors cheaply," he commented, "and one thing that Intel does very well is manufacture silicon." The relevance being that current optical device technology involves relatively exotic and expensive materials. "We are going to print silicon optical components," continued Demain. "That means we can do it cheaply and it means we have a new era of optics. This is the breakthrough." Key markets are likely to include consumer electronics, industrial and automotive – there could be as much as 130kg of copper in a high end car, which obviously has a significant impact on performance and energy efficiency. However, the first markets will inevitably be data centres, telecoms and high performance computing. Demain gave the example of the world's fastest computer Tiahne 2 in China, which features 3million Intel cores and consumes 17MW for operation and a further 7MW for cooling. The difference that silicon photonics could make to an application like this stems from the ability of the modules to sit right next to the CPU, rather than at the edge of the board, along with multiplexing of signals to use single fibres, cutting the amount of cabling by an order of magnitude. Demain expanded: "As bandwidth goes up, driving the signal to the edge of the board becomes a challenge – a 25Gbit signal going 9in creates problems with noise, routing, power, signal integrity. As speeds go beyond 25G, it becomes even more challenging. But we can now put the optics right next to the driving device and it becomes very clean, very short. This is the way optics will go in the long term." There are no plans on the horizon for Intel to do optical computing. While Demain admitted it was a possibility, he believes the costs are too high and the focus remains on fast, cheap and low power interconnect. "It has taken a decade of R&D – we went through the phase of creating the technologies, then through a phase of connecting them. Now, we are moving into the phase of launching them," he concluded.