VCSEL Technology Takes Off

By Kevin Fogarty and Ed Sperling

Vertical cavity surface emitting laser (VCSEL) technology, a proven but mostly niche technology until recently, is suddenly a very hot commodity thanks to the introduction of facial recognition in phones and other mobile devices.

VCELS primarily have been used as a low-cost way of tracking movement and transfering data in computer mice, laser printers and in fiber optics. But Apple’s decision to use VCSEL technology for facial identification in its iPhone X has pushed this technology in a different direction. That decision—and the flood of smartphone and consumer electronics makers who followed suit—forced the whole VCSEL market to scale up with new fabrication, testing and validation processes.

“In the VCSEL market, nothing happened for years,” said Ajit Paranjpe, CTO at Veeco. “But with new applications emerging, there has been a significant improvement in VCSEL technology. We’re finally getting over the learning curve to move this into high volume, which will be necessary when we get to the second big application, which is LIDAR where you have VCSEL arrays. Also VCSELs are already being used in server farms for rack-to-rack communication. After the adoption of board-to-board communication (such as optical backplanes),the Holy Grail is chip-to-chip communication, ultimately with an all-optical interconnect. The real question is when we will need it.”

VCSELs are just one of a handful of different silicon photonics approaches that are beginning to gain notice. Across the semiconductor industry there is work underway to bring one or more of these various technologies into the mainstream, particularly at advanced nodes where pushing electrons through wires is becoming much more difficult.

“With phones, there are lots of hidden and non-hidden cameras,” said Twan Korthorst, R&D director for photonic solutions at Synopsys. “Those cameras can identify whether you’re looking at the phone and take optical measurements, which is where VCSELs come into play. There is a detector, a light source and an image. VCSELs are a small discrete component.”

What is a VCSEL?
Put in perspective, VCSELs are just one of a number of possible chip-based light sources that can be used in these devices. What makes VCSELs so attractive is the laser is emitted perpendicular to the device. That has a number of benefits, from scaling to test.

“VCSELs can be tested at the wafer level using wafer probing and wafer test,” said Paranjpe. “With edge-emitting lasers, you have to dice the wafer and then build the rest of the device and test it, which makes it harder.”

That hasn’t slowed research into other areas, however, where much of the research is focused on stacking of different materials and integrating some or all of those into a package. What gets put into the package, or what resides outside the package, is still being worked on.

“There is some interesting work going on right now in the big foundries to create photonics elements that connect different die,” said John Ferguson, director of marketing for Calibre DRC applications at Mentor, a Siemens Business. “The main issue is that the wave guides have to be a certain dimension, which is a couple of microns. You have to keep a certain amount of space, which takes up a lot of area. Photonics doesn’t benefit from advanced nodes. In fact, 65nm is the most advanced node so far.”

Another opportunity lies in automotive applications like LiDAR with higher power requirements. This application will need to use larger VCSEL arrays.

“That will require purchasing more MOCVD (metal-organic chemical vapor deposition) systems to keep manufacturing capacity up to speed with demand for LiDAR systems,” said Mark McKee, product marketing director at Veeco. “Now the question is how to achieve the highest performance and highest productivity to meet market requirements. This is based on having the industry-leading MOCVD technology. Key requirements are uniform and laminar flow of the metal-organics and hydrides over the surface of the wafers, uniform and controllable temperature and sharp interfaces between layers. To achieve highest productivity, your platform will require long time between preventative maintenance cycles, fast recovery after preventative maintenance and fast epitaxial growth rates.”

Fig. 1:  Source: Finisar

VCSELs do their ranging and time-of-flight calculations using pulses of light with frequencies in the tens of gigahertz, identifying movement by looking at changes between one image and the next. It’s not clear if that approach could be adapted to improve on LiDAR’s tendency to use longer wavelengths and continuous scanning, noted David Hall, principal marketing manager at National Instruments.

“There doesn’t seem to be much overlap in requirements between a facial recognition system inexpensive enough to be integrated into a smartphone and a LiDAR system that’s supposed to scan continuously much further than VCSELs could manage,” Hall said. “It remains to be seen if either one could benefit from the approach of the other.”

LiDAR is a good potential market, but is less attractive than shorter-term automotive opportunities like motion-detection and face-recognition inside the cabin. The technology can be used to identify when a driver is drowsy, or allow passengers to control the infotainment or other systems with hand gestures, according to Craig Thompson, vice president of new markets for Finisar, which is supplies chips to Apple.

Fig. 2: Source: Yole Développement July, 2018 report

Finisar has been developing VCSELs for use in copper-to-fiber interfaces for carrier-grade data-networking equipment since 2004, when it bought the division that first commercialized VCSELs from Honeywell and expanded sales beyond computer mice and PC peripherals and into data networking.

But not everyone is on board with using VCSELs in their current form in LiDAR.

“VCSEL light tends to be close to the visible range, which can make them dangerous as you raise the power,” said Gilles Lamant, distinguished engineer at Cadence. “With LiDAR, you would need very high power to get any kind of range, which could be a risk to eyesight. Even close to the visible spectrum, though, they’re safe for low power applications, which is why people are looking at them for facial recognition and things like distance-ranging for cameras.”

Fig. 3: Source: Finisar

The real advantage of VCSELS is their convenience, flexibility and power, as well as their thermal efficiency compared to other laser sources, Lamant said.

“VCSELS emit light vertically, which makes it easy to build an array going perpendicular to the chip, and they’re quite stable with a pretty wide temperature variation,” Lamant said. “They’ve also been demonstrated sending data within one system where the coding of data signals within one system. That will run into limitations with the frequency you can do that, though so we’re still looking for a way to integrate optical with a silicon-based solution.”

VCSELs also deliver far more light per watt of power than alternative laser sources, Thompson said. Wiring is easier because electrical and thermal management can be done from underneath while the laser emerges from the top.

The amount of power VCSELs put out is also linearly scalable by size. Each laser aperture is independent and essentially identical. The more apertures on a chip, the more power it puts out. Connecting them all to a single power source causes them all to emit in unison. Wiring a chip into zones makes it possible to have them emit at different times and different patterns—the power output of which is determined by the number of apertures.

Fig. 4: Finisar high-power VCSEL in AIN cavity package (Source: Finisar)

“A data-networking VCSEL will have one aperture and one emission point on a die maybe a couple of hundred microns X and Y,” Thompson said. “In a high-power 3D sensing VCSEL there will be many hundreds of apertures and the die will be up to a millimeter square. It’s a very unique, nicely scalable laser structure. It’s almost a LEGO block kind of technology.”

It’s possible that thermal interference or other types of noise may hurt VCSEL usage in the datacenter, but tests able to identify that type of interference are uncommon, Ferguson said.

“This is not like an IC where you put it on a test bench with probes,” said Ferguson. “You need optical signal processing on the outside of these devices, and that’s not trivial. Some of the big systems companies are trying to drive the foundries to do that so they know what else they can do with this technology, because we’re starting to see more adoption beyond networking. There are applications for photonics in automotive with LiDAR and self-driving cars, and it’s looking like this technology is gaining traction. There are a number of companies that are exploring it, and several are now serious to the point where we expect products within the next year.”

There should be a lot of products coming out with VCSELs over the next year, Thompson said, but the volume of VCSELS being produced, tested and verified are all far higher than they’ve ever been, though it’s likely most outside the optical industry would realize the extent of the changes, Thompson said.

VCSELS were moderately successful for a decade in computer mice and other peripherals, which Honeywell commercialized in 1996, and have been popular as light sources in fiber-to-copper interfaces for carrier-grade data networking equipment since 2004. All of those were solid niche businesses, but were the were low-profile enough to keep VCSELs flying under the radar of most in the chip industry. After the iPhone X announcement, everything changed.

There were still some challenges in designing an oversized, overpowered VCSEL capable of bouncing 30,000 IR spots off a user’s face to assemble a 3D map quickly and accurately enough to power the Face ID authentication in iPhone X, according to Thompson said. But the biggest  challenge was figuring out how to meet the volume needs of the iPhone X. Apple announced that it would have to buy 10 times as many VCSEL wafers during the fourth quarter of 2017 as had been manufactured worldwide during the same period a year before.

This was why Apple gave Finisar $390 million in 2017. The goal was to turn an idle 700,000 square-foot facility in Sherman, Texas, into the “VCSEL capital of the US.” When it reaches full capacity later this year, the 700,000-square-foot former MEMC fabrication plant will turn out, all by itself, more VCSEL wafers by several orders of magnitude than the whole VCSEL industry was able to manage in the past, Thompson said.

Xiaomi and Oppo followed Apple’s announcement, promising to add 3D sensing functions to their mobile chips in mid 2018. That was followed by Huawei, Vivo and Samsung, which are expected to add VCSELs to some models this year, according to a market report from Yole Développement, which projects volume sales of VCSELs will rise from 652 million units during 2017 to more than 3.3 billion chips in 2023. Sales of VCSEL chips will increase from $165 million in 2017 to $3.1 billion by 2023, Yole predicts.

Philips Photonics, which boasts of having shipped more than 1 billion VCSELs compared to Finisar’s 300 million, spent 23 million euros in 2018 to double the capacity of its VCSEL facility in Ulm, Germany. And Austria-based ams announced it would spend $200 million to expand a VCSEL site in Singapore.

“The big story here is that the smartphone market has brought VCSELS into the mainstream, and companies like Apple have underwritten the cost of development and maturation of the technology necessary to scale to that king of volume manufacturing,” Thompson said. “Companies like Apple have underwritten much of the development and maturation of the technology for application in smartphones—and made us in the industry confident enough to invest in a big way to make VCSELs manufacturable in high volume.”

Before FaceID, standard VCSEL manufacturing was, almost exclusively based on MOCVD, which is commonly used for III-V materials to create polycrystalline thin films, and fab-efficiency measures like automated wafer testing and beam-imaging were “nascent and immature,” Thompson said.

“We’ve gone from 3-inch gallium arsenide wafers to 5-inch gallium arsenide. We’ve developed a more mature approach to automated wafer-scale tests throughout the fab process, which was very immature just a couple of years ago,” Thompson said. “We had to develop epitaxial wafers for these applications, scale the supply chain to keep up, develop new metrology and a new testing regimen based on learning borrowed from the RF industry to develop automated wafer tests and electrical probe. We had to develop, near-field and far-field optical testing to image the chip and its output up close and at a distance. We had to develop test methodologies to measure accurately things like number and function of apertures over a very large laser chip area and how the infrared light was being shaped and focused,” Thompson said.

The Sherman, Texas plant opened in July, 2018 and won’t reach full capacity until later this year, but already offers capabilities VCSEL makers could not imagine just a few years ago.

“The scale of the facility is orders of magnitude larger than anything the industry was producing collectively before,” Thompson added. “We don’t announce volumes, but we are getting close to the scale of a III-V RF foundry. We’re at a phase where we have working testers, proper near-field and far-field imaging testers that work and produce reasonable volumes at reasonable cycle times. We’re focused on doing multi up-testing and collecting a lot of that that allows us to do a lot of sample testing.”

Related Stories
LiDAR Goes Back To The Future
While the technology continues to be developed for automated driving, it is finding other applications
The Week In Review: Manufacturing
Apple’s VCSEL