Challenges In Photonics Testing

Alignment remains the top issue, but new developments could pave the way to high-volume manufacturing.

popularity

Photonics is poised for significant growth due a rapid increase in data volumes and the need to move that data quickly and with minimal heat. But to reach its full potential photonics will have to overcome several production hurdles.

The biggest challenge today involves alignment. While the industry is poised to produce billions of units, it still relies on testing practices that don’t scale.

“In photonics, you need to bring light in and have it coupled into photonic circuitry that is on the wafer,” Scott Jordan, head of photonics at Physik Instrumente USA (PI). “So the alignment has to be correct within a few tens of nanometers, with the exact angle and z position along the optimal axis to achieve optimal coupling — getting light in and out of a device or series of devices.”

Others agree. “Physical alignment is a big and important problem that a lot of companies are working on solving,” said Manish Mehta, vice president of marketing and operations for Broadcom’s Optical Systems Division. “If you’re working with high-density silicon photonics, you have to take optical elements and attach them to your optical engine in a way that you can get light in and out, and that’s a very common investment across the optics industry. Whether it’s with chip companies, transceiver companies, or now OEMs, everyone’s starting to invest in some of the capabilities for building that optical alignment as part of the manufacturing process.”

In fact, about 80% of the cost of a photonic integrated circuit is consumed in alignment processes, according to Jordan.

Like most things in photonics, it’s more complicated than CMOS manufacturing. “For photonics, testing means being able to couple and get information about light on the chip,” said Gilles Lamant, distinguished engineer at Cadence.  “It’s not like just having a landing pad on the electronics. On most chips, light goes parallel to the chip. Testing needs to pick it up and will alter it. Alignment is also for coupling, so in a package if you don’t have perfect alignment, you’re going to lose some power. You always hear people talking about how many dbs they’re losing on alignment. That’s where the challenge is.”

The tolerances involved increase the severity of that challenge. For example, a photonics interconnect is about 1,000X smaller than a typical on-chip interconnect.

“Photonics typically need sub-micron precision in placement of optical components within a package to align chip to fiber or through other optical elements,” said Robert Maher, CTO of Optical Modules and Coherent Solutions at Infinera. “This comes from the fact that the mode diameter in photonic ICs is on the order of few microns (1 to 3µm). The mode in fiber is larger than a typical waveguide device such as a laser or PIC and one needs a lens or series of precisely aligned lenses to map/align them. The mode size scales with the refractive indices and not just the size or diameter,” said Maher.

In many areas of photonics, there’s no margin, cautioned Twan Korthorst, group director of photonics solutions at Synopsys. “Depending on the tolerances, if your application is power budget-sensitive, and you lose a lot of dB in your fiber chip coupling or in your laser chip alignment coupling, then your whole system is not going to work.”

There’s another difference with electronics that underscores why it’s so important to get alignment right. “In electronics you can have an amplifier, so when you have a long chain of processing a signal, you can regenerate your signal along the way,” said Lamant. “In silicon photonics you can’t do that because there is no active device, so you have to be extremely careful about your insertion loss at each level. Of course, in electronics you also try to mitigate your losses, but it’s fairly common to insert a buffer to regenerate your signal. That is very hard to do today in photonics.”

There are many research efforts dedicated to grafting active devices on top of silicon. Still, photonics has its own idiosyncrasies.

“It is important to consider the dramatic differences in an electrical channel versus an optical channel,” said John Calvin, senior strategic planner for IP wireline technology at Keysight Technologies. “The ‘channel’ or trace between two electrical chips may only be a few centimeters, but the attenuation can be on the order of 40 dB. An optical fiber has extremely low loss (<0.5 dB/km) and channels may be 10 km or more in length. Rather than frequency-dependent attenuation, optical transmitters are tested in the context of signal dispersion and expected receiver sensitivity. Optical signals ride on an optical carrier, so the carrier performance in terms of wavelength is also important to control in order to avoid channel interference and manage dispersion.”

Testing options
Consider what needs to happen with transmission, especially as data rates increase. “A transmitter made of discrete components requires alignment of optical components to minimize attenuation,” Calvin said. “Optical modulators also require polarization alignment to operate correctly. While not anything like optical beam alignment, there is also the issue of optimizing the electrical to optical conversion or modulation of the light.”

In this scenario, alignment of optical beams can be completed with simple power measurements. “Optical modulation quality must be assessed with high-speed instrumentation,” Calvin said. “For intensity-modulated direct detect systems, the transmitter eye diagram acquired with a digital communications analyzer is assessed and used to verify eye-diagram quality. For coherent (complex modulation) systems, the constellation diagram is measured using an optical modulation analyzer.”

In general, there are two kinds of tests to determine alignment — active testing, which predominates, and passive testing.

Active alignment is a feedback loop. “People bring in a fiber and power it up, and while moving around the fiber with an XYZ and even rotation system, they measure the actual coupling from the fiber to the chip, and fix it, when optimal transmission is measured,” said Synopsys’ Korthorst. “So then the test is actually part of the actual assembly process, because you have a feedback loop where you measure while you fix by glueing or soldering.”

At this point in photonics development, active testing is either done by a company creating its own in-house testing platform, or through a third-party testing tool vendor. Jordan’s company, PI, claims the fastest alignment speed for test and assembly in the industry, a bedrock for high-volume manufacturing, and likely to remain a key differentiator among vendor claims.

“Over the last 10 years, there are more standard solutions for testing and evaluation of optical sub-assemblies and packages,” said Michael Lebby, CEO of Lightwave Logic. “That’s going to keep increasing, because as volumes keep increasing in the optical industry, we have to get the testing time down. When I first started back in 2000, the time it took to test and evaluate an optical module was two to three hours. Now it’s down to under an hour and where we’re going is to get it done in a few minutes.”

Broadcom has announced a passive testing system, which if it performs as promised, should greatly increase the speed of testing and support high volume manufacturing of co-packaged optics (CPO). Passive testing eliminates the feedback loop and is a self-contained, self-diagnosing system.

For CPO with detachable fiber, there will still be active alignment steps required to attach the passive optical components to the optical engine. However, once engines are fully assembled, they can be tested at both engine and CPO level by passively inserting fiber cables, which enables a robust test flow at various levels of the assembly. “You don’t attach fiber permanently. You have an optical engine that you have to insert a fiber cable into in order to get your light in and out, but it is detachable,” said Mehta. “Because of that, we can build test systems where you just insert your optical engine, and you insert the fiber, and you’re able to do test and it’s passively aligned, it’s self-aligned. It’s part of our manufacturing process that you get the alignment.”

Greater traction
Much of this is a sign that photonics is coming of age. Until recently, its small market size may have offered less enticement for larger tool vendors to join in with solutions.

While photonics has long been established in telecom, it’s a market that requires fewer installations than newer applications, like automotive lidar. “The volumes haven’t been that high because most of these high-speed optics are used in core bulk massive transmission applications,” said Rob Shore, senior vice president of marketing at Infinera. “Only recently are we starting to get more toward edge applications and pluggables, where the volumes go from hundreds or thousands of units to 10 million-plus units.”

Photonics was a big buzzword in the 1990s, as the promise of the early internet drove telcos to install fiber-optic cables. But once enough fiber was in place, the excitement receded and the industry merely hummed along. Now, with the recent data boom, the excitement has returned and the industry is partying like it’s 1999, with one important exception — this time the party isn’t likely to sputter out.

“The magic word is consumer, because that’s when you start getting into volumes of billions, instead of a couple millions,” said PI’s Jordan. “The current interest in silicon photonics has been driven by energy efficiency, capacity, and speed for data centers. It’s what’s going to drive this field from now on. That’s the difference compared to what happened in 1997 to 2000, when there was a single application.”

Changes are on the horizon today as the industry needs are shifting, particularly in data centers. “The last several generations of data center specifications have relied heavily on dealing with channel loss through re-timing and signal regeneration,” said Keysight’s Calvin. “BER was the priority with less of an emphasis on latency and aggregate power performance. The resulting DSP-heavy and power-hungry architectures are under serious scrutiny today. There are direct drive (sometimes called linear drive) initiatives that eliminate many of these regenerative circuits and emphasize linear/analog signal conditioning. In this direct drive scenario, what were once two clearly separated domains become inextricably tied together as part of single system. Test strategies will need to be based on maintaining the information signal integrity as it flows from electrical to optical and back.”

Conclusion
Industry maturity will likely change some, but not all of the testing environment.

“With the advent of silicon photonics, much of the manufacturing variability drops out as the ‘system’ is built using conventional, microelectronic CMOS type fabrication processes,” said Calvin. “This can simplify or even remove some of the tests required. But process variability still exists, and some testing will still be required, such as the waveform quality at high data rates.”

Overall, there is optimism that greater demand will lead to appropriate equipment becoming more widely available, in a familiar industry cycle. “Photonics is just like the semiconductor industry was 35 years ago,” said Jordan. “It’s just people all over the place doing little manual jobs using custom machines. But that ecosystem grew, and today you can equip a semiconductor fab just by writing purchase orders and we see that happening in photonics as well.”

Mehta agreed, highlighting changes in a different part of the industry. “About a decade ago, folks started going to some of the high-precision, die-attach vendors and asking for customization on tools that at the time were being used for non-optical products. Over the next two to three years, there was a fair amount of customization required, but then it was deployed across 10 million to 20 million units a year of optical transceivers. Now, if you walk into any of those vendors, and you ask them for the kind of alignment requirements that are needed for optics, they know exactly what you’re looking for. We just have to get the volumes.”

 

Related Reading
Transitioning To Photonics
High speed and low heat make this technology essential, but it’s extremely complex and talent is hard to find and train.
Standards: The Next Step For Silicon Photonics
More data and denser designs are opening the door for photonics.



Leave a Reply


(Note: This name will be displayed publicly)