From the world’s fastest supercomputers to your laptop or phone, almost all modern devices still rely on electric wiring to transmit data – a technology dating back to 19th-century telegraphs. Cheap and reliable, copper wire has linked ever larger quantities of transistors and support circuitry on ever smaller chips for decades.
Yet as we rapidly approach the physical limit of chip miniaturisation, the shortcomings of copper wire come into focus. Limited bandwidth, current leakage and crosstalk between adjacent wires produce bottlenecks for computing innovators. There are also practical concerns like the space copper wires take up, the power they consume and the heat they emit.
One solution has been in light-based technologies, like optical-fibre cables, which use photons to transmit data at much higher speed and bandwidth. But the exorbitant cost of replacing electric wiring in integrated circuits with cutting-edge photonics has proved to be a roadblock to commercialisation. What is needed is a photonics technology that can be made using existing high volume, inexpensive manufacturing methods.
Enter silicon photonics. Its great advantage is that it can be manufactured just like normal computer chips but with patterned silicon to transmit data-carrying laser signals. Able to carry more data while consuming less power and without heating up or causing any degradation in signal, silicon photonics will disrupt a range of technologies. These include data interconnects, supercomputers and sensors to diagnose disease, monitor the environment or even detect explosives.
An imperfect fit
Although silicon photonics is a perfect solution on paper, for more than 30 years scientists have struggled to surmount its various shortcomings. While silicon is opaque in the visible spectrum, it is transparent at infrared wavelengths used in optical transmission, so it is good at guiding light. Where it falls down is in emitting light. Silicon photonics expert Lars Frandsen from the Technical University of Denmark (DTU) gets into the technical details: “The electronic bandgap of silicon makes it transparent in the telecommunication wavelength range, but it is an in-direct bandgap, which make silicon a very poor emitter.” This is why the lasers that drive optical communications are made of more exotic materials such as indium-phosphide (InP) and gallium-arsenide (GaAs).
Another barrier comes when, if you do manage to get silicon to emit a light signal, you mix it with other signals to perform computation. Unfortunately, silicon is also very bad at this task. “Silicon suffers from large signal losses at high pump powers, i.e. where many photons are present in the silicon waveguide,” adds Frandsen.
The Walter Schottky Institute of Technical University Munich
At the Walter Schottky Institute of Technical University Munich, physicists develop a process to deposit nanolasers directly onto silicon chips.
Silicon photonics today
Despite these shortcomings, some silicon photonics devices are coming to market. California’s Luxtera stole a march on its competitors when it started shipping its optical interconnect products in 2008. Replacing some copper connections, these devices are tiny silicon chips with built-in InP lasers and silicon waveguides that can shuttle data at up to 100 Gbps between computers via optical fibres just a few millimetres thick. “Data centre operators are working feverishly to supply the networking bandwidth to support the world’s insatiable demand for data,” says Luxtera’s Ron Horan. “Luxtera has shipped more than 1 million optical transceivers, including 100 Gbps solutions in very high volume.” Tech giant Intel brought its technology to market in 2016. With IBM, Cisco and others working on similar technology, there is little doubt that optical interconnects are the silicon photonics technology that will have the first big impact.
Yet switch-to-switch optical connections are just the beginning. Many see silicon photonics inevitably getting deeper and deeper into computers as the technology matures, allowing supercomputers to reach exascale speed (1018 calculations per second) around the human brain’s processing power at the neural level. In 2015, US researchers described in a Nature paper how they had demonstrated the first photonic-electronic hybrid processor that uses light for ultrafast communications. Packing two processor cores with more than 70 million transistors and 850 photonic components onto a 3 x 6 mm chip, it was the first to integrate photonic interconnects, or inputs and outputs (I/O) required to talk to other chips. Staggeringly, the chip had a bandwidth density of 300 Gbps per mm2, a 10- to 50 fold improvement on electronic-only microprocessors.
The achievement did however come with a caveat: the team had to use externally generated laser light to illuminate the chip. Major advances still need to be made in terms of developing and integrating on-chip lasers before commercialisation. Until recently, laser light sources could be attached to silicon only in complicated and elaborate manufacturing processes. But now Technical University of Munich physicists Gregor Koblmüller and Jonathan Finley have made a significant breakthrough, developing a process to deposit nanolasers directly onto silicon chips. The team evaporated a thin layer of silicon oxide on the silicon wafer, then etched tiny holes in it. Next, they grew freestanding GaAs nanowires out of these holes. The result? Nanowire lasers 1,000 times thinner than a human hair. Says Koblmüller: “Our approach is novel and shows great site-selective integration of nanolasers on a silicon platform.”
Data centres and the IT industry will not be the only beneficiaries of silicon photonics. CARDIS (Early stage CARdio Vascular Disease Detection with Integrated Silicon Photonics) is an EU project exploring the use of silicon photonics to identify people at risk of cardiovascular diseases. The aim is to build a mobile, low-cost device similar to a handheld supermarket scanner, able to assess the heart’s vital signs with one click of a button. The project hinges on a multi-beam silicon photonics integrated laser vibrometer – a scientific instrument used to make non-contact vibration measurements of a surface – to easily target and assess superficial arteries. “An electronic chip could not do the job since it does not allow for contactless detection of skin movement at the nanometre level,” explains Project Coordinator Roel Baets from Ghent University. “The silicon photonics integrated chip holds the entire optical functionality of the laser Doppler vibrometer.” Baets sees a number of other health applications for the technology. “There’s screening and continuous glucose monitoring for diabetes patients, and there is lots of potential in diagnostic assays, sensing of ions in sweat and breath analysis,” he explains. All these applications boil down to integrating photonics sensors in silicon, thus many other devices could be engineered, like Lidar sensors used by autonomous vehicles or chemical sensing for environmental monitoring. “The Internet of Things will require billions of distributed sensors everywhere in our society,” adds Frandsen. “Silicon photonics will be a strong candidate for making this possible.
Rivals challenge silicon’s supremacy
Developed over more than five decades, silicon would seem to be the only option when it comes to future photonics-enabled chips. But leading lights at European universities disagree.
More exotic materials with far better photonic properties than silicon are being explored. Eindhoven University of Technology’s Kevin Williams studies applications that add indium-phosphide (InP) photonic elements on top of silicon electronic elements. “InP technology enables semiconductor light sources and optical amplifiers at near-infrared wavelengths unlike other integration technologies,” he says.
Already a billion-dollar industry, InP integrated photonics is the preferred solution for high performance telecommunications. The material’s properties and the maturity of the technology enable the creation of circuits with lasers, amplifiers, modulators, filters and detectors at high density. Only cost and size hold back InP photonics from moving to new applications beyond telecommunications. “Performance-wise, the InP material is superior to silicon as it holds an electro-optic effect as well as a direct band gap,” adds DTU’s Lars Frandsen.
One of many scientists working at the cutting edge of photonics research, Frandsen is more interested in a different material. His team dopes silicon nitride (Si3N4) with even more silicon to create silicon-rich nitride (SiNx). Regular Si3N4 has a bandgap five times larger than silicon, meaning waves can be better guided and preserved. However, this comes at the cost of fewer and less useful effects produced from the light’s interaction with the material it is travelling through (known as nonlinear effects), thus requiring higher pump powers and limiting the waveguide design. With SiNx, Frandsen’s team has been able to increase the nonlinear parameter 10-fold in SiNx waveguides compared to Si3N4.