FEATURE: PHOTONIC INTEGRATION
in speed, latency and power consumption in solving intellectual tasks that are unreachable by conventional digital electronic platorms. ‘Tis is perhaps the first real-world
application of photonic neural networks, and shows how the system can process optical signals in real time – that is, predicting and compensating for fibre nonlinearities in an over 10,000km trans-pacific submarine transmission link,’ she added.
Alternative hardware solutions In view of the fact that today’s network traffic is ‘growing at a rate of 30 per cent annually’, Huang stressed that an increase in optical communications capacity of more than an order of magnitude ‘will be needed within the next decade to sustain the continued operation of the internet’. Moreover, signal distortions caused by detrimental effects in optical fibres ‘become more severe in high-capacity and long-distance communication systems’. Huang explained ‘Digital signal processing
(DSP), with its proven ability to combat transmission distortions, is a key technology that enabled the growth of the internet in past decades. However, the advances in high-speed DSP implemented CMOS electronic chips were heavily reliant on Moore’s Law scaling. Today’s semiconductor technology has approached the limits in terms of power dissipation, density and feasible engineering solutions. DSP capacity will therefore find it increasingly challenging to support the continued exponential growth of internet traffic in the post-Moore’s Law era,’ said Huang. In recognition of this limitation, Huang
pointed out the key objective of the research is to develop alternative hardware solutions that can address the potential optical network capacity crunch caused by the slowdown of Moore’s Law – prompting the team to use a neural network implemented in hardware on an integrated photonic chip enabled by silicon photonics, which emulates the underlying neural network model with photonic devices and circuits. When compared with existing approaches,
Huang observed that a key advantage of using the neural network for optical communications is that ‘a simple photonic chip with just a few neurons and tens of devices can, in principle, outperform commercial DSP chips in throughput, latency and energy use. ‘In addition, our photonic neural networks
make use of high-quality waveguides and photonic devices – that is, photodetectors and modulators – that are originally designed for optical communications. As a result, the photonic neural networks can always accommodate fibre communication rates and enable real-time processing for future optical networks, promising to address the potential optical network capacity crunch caused by the slowdown of Moore’s Law,’ she said.
Seamless interface Although the current work is focused on addressing the signal distortions in a single wavelength channel, Huang points out that
www.fibre-systems.com @fibresystemsmag
we are working with NEC to figure out the commercialisation plans,’ she added.
DSP CAPACITY WILL FIND IT INCREASINGLY CHALLENGING TO SUPPORT THE CONTINUED EXPONENTIAL GROWTH OF INTERNET TRAFFIC IN THE POST-MOORE’S LAW ERA
the photonic neural network can also interface ‘seamlessly’ with WDM optical fibre systems. Moving forward, the research team plans to use this novel architecture to process multiple WDM channels in parallel and in the optical domain. According to Huang, this would result in a bandwidth increase ‘over Terahertz, significantly beyond the capability of DSP’. ‘Tis unique feature helps with inter-
channel, nonlinear compensation in a WDM communication system, which DSP struggles with, while offering low-energy operation by eliminating power-hungry ADCs – which may consume more than 40 per cent of the energy in some transmission systems,’ she said. ‘Meanwhile, photonic neural networks,
with their unique advantages in bandwidth, latency and power consumption, can play a unique role in enabling many new domains of applications – for example, machine learning acceleration, nonlinear programming [and] intelligent signal processing. Our future research will explore more applications of our photonic neural networks and
Indium phosphide Another interesting ongoing initiative is Infinera’s work on the development of high- performance indium phosphide (InP) as a platorm for PICs for optical communications. Building on its reputation as an early pioneer in the field of photonic integration that built the industry’s first large-scale PIC in 2005, the company is now focused on leveraging high- InP-based PICs that integrate a wide range of optical functions on a single chip. As Paul Momtahan, director – solutions marketing at Infinera, explained, this approach ‘reduces cost, footprint and power consumption while improving performance and reliability. ‘We have also invested heavily in building
our own state-of-the-art InP PIC fab – the only equipment manufacturer to have done so to date. In addition to the obvious cost advantages of this vertical integration, it enables fast redesigns for optimised performance, tight quality control and a faster ramp to volume for new technologies and products, such as ICE6,’ he said. In Momtahan’s view there are a number of
inherent advantages to using Indium Phosphide in PICs and optical communications systems. To begin with, in terms of the modulator, he points out that InP’s electro-optic effect is ‘fundamentally more efficient than the plasma dispersion effect used by silicon photonics’ – and produces a phase change ‘up to 10 times higher for a given unit length and voltage, resulting in more compact and power-efficient modulators with lower loss, beter linearity, larger modulation voltage for higher transmiter signal to noise, and therefore beter reach.’ ‘For this reason, as far as I am aware, all fiſth-
generation 90+ GBd coherent optical engines currently on the market use an InP modulator,’ he said. Momtahan also observed that InP can
support gain elements, such as laser and amplification – semiconductor optical amplifiers – enabling what he describes as ‘more monolithically integrated photonic
A relative comparison of modulator bandwidth Issue 35 n Spring 2022 n FiBRE SYSTEMS 17
nikkytok/
Shutterstock.com
Lightwave Logic
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40
