The quiet period of silicon photonics
Michael Hochberg discusses his book on silicon photonics and the status of the technology. Hochberg is director of R&D at Coriant's Advanced Technology Group. Previously he has been an Associate Professor at the University of Delaware and at the National University of Singapore. He was also a director at the Optoelectronic Systems Integration in Silicon (OpSIS) foundry, and was a co-founder of silicon photonics start-up, Luxtera.
Part 2: An R&D perspective
If you are going to write a book on silicon photonics, you might as well make it different. That is the goal of Michael Hochberg and co-author Lukas Chrostowski, who have published a book on the topic.
Michael HochbergHochberg says there is no shortage of excellent theoretical textbooks and titles that survey the latest silicon photonics research. Instead, the authors set themselves the goal of creating a design manual to help spur a new generation of designers.
The book aims to provide designers with all the necessary tools and know-how to develop silicon photonics circuits without needing to be specialists in optics.
“One of the limiting factors in terms of the growth and success of the field is how quickly can we breed up more and more designers,” says Hochberg.
The book - Silicon Photonics Design: From Devices to Systems - starts by exploring the main silicon photonics building blocks, from optical waveguides and grating couplers to modulators, photo-detectors and lasers. The book then addresses putting the parts together, with chapters on tools, fabrication, testing and packaging before finishing with system design examples.
The numerical tools used in the book are mostly based on the finite-difference time-domain method, what the authors describe as the typical workhorse in silicon photonics design. Hochberg admits that the systems software tools, in contrast, are less mature: “It is a moving target that will change year to year.”
Myths
Hochberg is also a co-author of a Nature Photonics’ paper, published in 2012, that debunks some of the myths regarding silicon photonics. “We wrote the myths paper after seeing an upswing in the ratio of hype-to-results going on,” says Hochberg.
He says part of the problem was that people were claiming silicon photonics was going to solve problems that it was plainly unsuited to address, for example integrating photonics with cutting-edge ultra-scale sub-micron electronics, for instance at 16 nm and 28 nm nodes. “That is not a practical solution for any near term problem,” says Hochberg.
More recent events, such as Intel’s announcement in February that it is delaying the commercial introduction of its silicon photonics products, highlights how bringing the technology to market is a significant engineering challenge. Instead, we are in a quiet period for silicon photonics, he says. Companies are getting into serious product mode, where they stop publishing and start focussing on building a product.
Moreover, these products - what he refers to as second-generation silicon photonics designs - are increasingly sophisticated with more functions or channels placed on the chip. “It is the standard story of almost any technology in silicon,” he says. “Silicon wins when you can do more stuff on a single chip.”
Silicon photonics and III-V
Hochberg stresses that while it is an understandable desire, it is very hard to compare the performance of silicon photonics as a whole with traditional optical components using III-V compounds. The issue being that silicon photonics comprises many different platforms where designers have made tradeoffs. The same applies to III-V compounds where there are hundreds of processes aimed at thousands of different products. “It is very hard to compare them in a generic way,” he says.
“The great advantage silicon photonics gives you is access to first-rate fabrication infrastructure,” says Hochberg. Silicon photonics offers 8- and 12-inch wafers, high volume foundries, high process control, the ability to ramp to high volumes and achieve high yields of complex-structure designs with hundreds, even thousands of components on-chip.
In contrast, III-V materials such as indium phosphide and gallium arsenide offer higher mobilities - electrons and holes move faster - and, unlike silicon, can straightforwardly emit light.
“The downside is that III-V foundries use technology processes that silicon stopped using 20 to 30 years ago,” says Hochberg. Wafers that are 2-, 3- or 4-inch in diameter, lithography that is ten times coarser than is used for silicon, process controls that are less advanced, and less automation.
If you are going to design a complex chip with lots of different components that require a predictable relationship with each other, this is where silicon tends to beat III-Vs, he says.
But the claim of large silicon wafers and huge volumes is what silicon photonics proponents have been promoting for years, and which has fed some of the false expectation associated with the emerging technology, says one industry analyst.
Hochberg counters by highlighting two trends that play in silicon photonics’ favour.
One is the well-known one of optics slowly replacing copper. This has been going on for 40 to 50 years, he says, in long haul, then in metro and now linking equipment in the data centre. “This will continue for shorter and shorter distances and then, at some point, stop,” he says. That said, Hochberg stresses that there are other applications for silicon photonics besides data communications.
“Just because you run out of opportunities at shorter and shorter reach at some point in the distant future, doesn't mean that the field collapses,” he says. “There's a lot of other cool stuff being done in silicon photonics these days with serious commercial potential.” Example applications include medical and remote sensing.
Once you can do something in silicon and do it adequately well, it tends to displace everything else from the majority of the market
The second trend he highlights is that silicon ends up dominating fields, not necessarily because it is the best choice in terms of performance but because it ends up being so cheap in scale. “Once you can do something in silicon and do it adequately well, it tends to displace everything else from the majority of the market.”
There are up-front costs of getting silicon photonics into a CMOS fab so companies have to be judicious in choosing the applications they tackle. “But once the infrastructure gets going to make a new application, the speed with which the industry can scale is just mind-blowing,” he said.
At Coriant, Hochberg leads a team that is doing advanced R&D. “We are doing advanced research with the goal to develop new technology that may eventually make its way into product.”
Does that include silicon photonics? “There is certainly an interest in silicon photonics; it is one of the things we are exploring,” says Hochberg.
Further reading:
Book: Michael Hochberg and Lukas Chrostowski, Silicon Photonics Design: From Devices to Systems, Cambridge University Press, 2015
Paper: Myths and rumours of silicon photonics, Nature Photonics, Vol 6, April 2012.
Silicon photonics economics set to benefit III-V photonics
Silicon photonics promises to deliver cheaper optical components using equipment, processes and fabrication plants paid for by the chip industry. Now, it turns out, traditional optical component players using indium phosphide and gallium arsenide can benefit from similar economies, thanks to the wireless IC chip industry.
Valery TolstikhinSilicon photonics did a good thing; it turned the interest of the photonics industry to the operational ways of silicon
So argues Valery Tolstikhin, head of a design consultancy, Intengent, and former founder and CTO of Canadian start-up OneChip Photonics. The expectations for silicon photonics may yet to be fulfilled, says Tolstikhin, but what the technology has done is spark interest in the economics of component making. And when it comes to chip economics, volumes count.
“For III-V photonics - indium phosphide and related materials - you have all kinds of solutions, designs and processes, but all are boutique,” says Tolstikhin. “They are not commercialised in a proper way and there is no industrial scale.” The reason for this is simple: optical components is a low-volume industry.
This is what Tolstikhin seeks to address by piggybacking on high-volume indium phosphide and gallium arsenide fabrication plants that make monolithic microwave integrated circuits (MMICs) for wireless.
“To take photonics out of boutique fabs, you need to do some standardisation and move to a fabless model, then you can load the fabs day and night with wafers,” says Tolstikhin. “That is the only way to make a process mature, reproducible and reliable.”
Tolstikhin has spent the last decade pursuing this approach. “The idea is to use something available in indium phosphide which is relatively close to a pure-play foundry.” A pure-play foundry is a fab that makes chips but does not design, market or sell them as its own products.
Tolstikhin’s first involvement was at start-up OneChip Photonics which developed an indium-phosphide platform that used a variety of photonic devices to make photonic integrated circuits (PICs), based on a commercial MMIC process.
The issue with III-V integrated photonics is that to implement different functions - a passive waveguide and a laser, for example - different materials are needed. “What makes a low-loss passive waveguide, does not work for the laser,” says Tolstikhin.
To overcome this, the wafer is repeatedly etched in certain areas, to remove unwanted material, and new layers grown instead with the required material, a process known as selective-area etch and regrowth. This is a complicated and relatively low-yield process that is custom to companies and their fabs, he says: “This is how all commercial lasers and PICs are made.”
In contrast, MMICs using indium phosphide do not need regrowth, simplifying the process considerably. To use a MMIC fab for an optical design, however, it must be developed in a way that avoids the need for regrowth stages.
“At OneChip we believe we did the first commercial laser - not just the laser but the PIC with it - regrowth-free,” says Tolstikhin. “It was made in a MMIC fab, that is the key.”
“To take photonics out of boutique fabs, you need to do some standardisation and move to a fabless model, then you can load the fabs day and night with wafers”
Wafer economics
To understand the relative economics, Tolstikhin compares the number of wafers - wafer starts - processed in silicon, indium phosphide and gallium arsenide.
One large TSMC fab has 400,000 12-inch CMOS wafer starts a year whereas globally the figure is equivalent to some 70 million such wafers a year. For MMICs, one fab Tolstikhin works with has 15,000 4-inch indium phosphide wafer starts a year whereas a large optical component company uses just a couple of thousand 3-inch indium phosphide wafers a year.
“In photonics, the [global] volumes – even for components going into the most massive markets like PON and the data centre interconnects – are still very low,” says Tolstikhin.
Gallium arsenide is somewhere in between: Win’s fab in Taiwan, which makes power amplifiers for wireless and other MMICs, has 250,000 6-inch wafers starts a year, while TriQuint’s fab in USA, with similar product line in wireless, totals 150,000 6-inch wafer starts a year.
Such volumes are not negligible and exceed all the needs of photonics, he says, enabling photonics to make claims similar to those trumpeted for silicon photonics: a mature process with a well-established quality system and, with its volumes, delivers better economics.
Moreover, if applications that currently are based on indium phosphide could be transferred to gallium arsenide, that would give an order of magnitude economies of scale, says Tolstikhin: “One example is mid-reach single-mode optical interconnects with an operating wavelength around 1060 nm, with gallium arsenide used for the transmitter, receiver and transceiver PICs”.
And while the scale of III-V semiconductor manufacturing may still be much lower than CMOS, the up-front cost involved in using a III-V fab is also much less.
Using III-V semiconductors for analogue electronics like the laser /modulator drivers or the trans-impedance amplifier also delivers a speed advantage: heterojunction bipolar transistors (HBTs) in indium phosphide have been demonstrated working at up to 400 GHz, and these, being vertical devices, do not have their speed scaled with lithography. In contrast, CMOS analog electronics is much slower and its device speed is scalable with lithography resolution. A 130 nm CMOS process, the starting point for silicon photonics, cannot support optical components with bit rates beyond 10 Gbps.
Design house
Intengent, Tolstikhin’s company, acts as a bridge between OEMs building optical components and sub-systems and the III-V foundries making photonic chips for them.
He compares Intengent to what application-specific IC (ASIC) companies used to do for the electronic chip industry. Intengent works with the OEM to specify and design the photonic chip based on its system application and then works with the fab to develop and turn the chip into a product by meeting its design rules and process capabilities.
“The aim is that you can go and design within existing fabs and processes something that meets the customer’s application and requirements,” he says.
Tolstikhin is also working with ELPHiC, a Canadian start-up that is raising funding to develop single-mode mid-board optics. The indium-phosphide design combines analogue electronic circuitry with the photonics.
“It appears the best way [to do mid-board optics] is based on electronic and photonic integration onto one substrate and indium phosphide is a natural choice for such a substrate,” he says.
Tolstikhin makes clear he is not against silicon photonics. “It did a good thing; it turned the interest of the photonics industry to the operational ways of silicon: standardised processes, pure-play foundries, device designs separate from the semiconductor physics, and circuit designs separate from the wafer processing.”
As a result, something similar is now being pursued in III-V photonics.