Optical components

Intel Detects More Silicon Photonics

It's been almost two years since Intel Corp. (Nasdaq: INTC) announced any "breakthroughs" in silicon photonics. Well, we can't let that stand, can we?

Fear not. Intel is publishing its latest results today in the journal Nature Photonics, describing an Avalanche Photodiode (APD) -- a type of detector for the receiving end of an optical link -- that's made of silicon (but not completely -- there's also a germanium layer).

Intel's silicon photonics efforts so far have focused on modulators and lasers. (See Intel at 40, Intel Pushes Silicon Modulator , Intel Fires Up Silicon Laser, Intel Claims Laser Breakthrough, Intel Gets Passive on Optics, and Intel's Modest Modulator .)

The APD is new ground, not just because it's a different part, but because its performance outdoes "any equivalent device in a III-V-based or exotic material," says Mario Paniccia, an Intel fellow and director of the company's Photonics Technology Lab. (III-V, or "three-five," refers to a class of compounds such as indium phosphide (InP) or gallium arsenide.)

That's a first. Silicon photonics have always been a tradeoff. The devices would be easier to integrate and cheaper to manufacture, since they can be built using complementary metal-oxide semiconductor (CMOS) techniques that are commonplace in the chip world -- but the performance suffers. Intel has been aiming for CMOS devices with 90 percent of the performance of InP ones.

The APDs aren't anywhere close to being a marketable product, by the way. "This is a research result. It's actually a very new result, Paniccia says.

Like any big company, Intel has started and ended its share of projects. (See Will Intel Trash Telecom?, Marvell Takes a Bit of Intel, Intel Dumps Dialogic, and Intel Hands Off to Cortina.) But silicon photonics has stuck, a testament to how vital Intel believes this technology is to the future of computing.

There have been changes, though. All the photonics work is now being done at Numonyx BV , a spinoff created early this year out of the Intel and STMicroelectronics NV (NYSE: STM) flash memory businesses.

Why? It so happens, Intel's silicon photonics work was being done at what is now Numonyx's fab. Intel found it easiest to just keep the operation in place, Paniccia says. So, the engineers technically work at Numonyx and build their devices on the same Numonyx production lines that are churning out high-volume memory chips.

Meanwhile, silicon photonics are reaching the commercial stage, mainly in the form of active optical cables for data centers. Kotura Inc. , Lightwire Inc. , and Luxtera Inc. are among the companies producing or pursuing commercial products. (See Silicon Photonics Advance PIC Possibilities, Luxtera Goes Commercial, and Lightwire Debuts Its Silicon Photonics.)

Going the distance
Intel's silicon photonics efforts are aimed mostly at short-reach connections, but the APD could easily be applied to a telecom network. The devices usually get mentioned in the context of long-haul spans, partly because they're too expensive to use elsewhere -- $200 to $300 apiece, Paniccia says.

The advantage of an APD is that a weaker light source can generate a sufficient current. That means you can take some liberties on the transmission side -- moving the source a farther distance away, for instance. Among the possible applications Paniccia cited was the fiber-to-the-home network, where APDs could conceivably be used to extend the reach of fiber links.

Performance for APDs can be measured in the gain-bandwidth product -- that is, the device's gain multiplied by the speed of the connection, which comes out to a fixed number measured in Hertz. (Note that this means the gain goes down as the bandwidth gets faster.)

For an indium phosphide APD, that gain-bandwidth product is around 120 GHz, Intel says. Intel's silicon APD is showing 340 GHz, implying that it would have better gain than InP devices.

Intel didn't specify the speed it's aiming for with APD, but the company is shooting high with its marketing, saying a silicon APD could be an aid in 40-Gbit/s networks. That would be quite a leap, as APDs are only available in speeds up to 2.5 Gbit/s today.

"A 40-Gbit/s APD might be really pushing it, but as something they're talking about for the future, it might be reasonable," says Ali Abouzari, vice president of sales for CyOptics Inc.

To describe which part of the APD is made of silicon, it's helpful to look at how an APD works. A normal photodiode receives a photon of light and produces an electron/hole pair (you can think of a "hole" as the opposite of an electron), creating electrical current. An APD adds a multiplication region where that reaction gets amplified, creating many more electron/hole pairs and a stronger current.

Intel used silicon for the multiplication region. But to absorb the photon and get the process started, Intel needed germanium, because silicon is transparent to the infrared wavelengths used in communications. Silicon can't "catch" the light.

Plenty of challenges exist with this approach. One is that the silicon and germanium atoms form lattices that don't quite match up, and that can cause some current to leak out even when there's no light present. Intel is still working on getting that "dark current" down, Paniccia says.

— Craig Matsumoto, West Coast Editor, Light Reading

nodak 12/5/2012 | 3:25:43 PM
re: Intel Detects More Silicon Photonics "That would be quite a leap, as APDs are only available in speeds up to 2.5 Gbit/s today. "

I think you might want to review this statement. Just a quick search turns up APDs being used in 10G receivers in 1996 experiments and in 2001 products (not an exhaustive search, probably things that could turn up earlier). Were you perhaps talking about the use of PIN diodes?
Pete Baldwin 12/5/2012 | 3:25:43 PM
re: Intel Detects More Silicon Photonics Luxtera already has silicon APDs that I think it's shown publicly. It's at least talked about them publicly, in papers at OFC/NFOEC, ECOC, and other shows.

It's the same concept Intel is using -- an absorption layer made of Ge and a multiplication layer of Si -- but Luxtera says its model is a waveguide-based photodetector.

Intel, by contrast, hasn't gotten its APD to work in waveguide form yet; on the conference call, the company says it's still working on such a thing.
litsem 12/5/2012 | 3:25:42 PM
re: Intel Detects More Silicon Photonics I may be wrong but I believe Germanium APDs in Silicon have been done in lab and even commercialized by companies like Luxtera long before this article. I understand it's a first for Intel and maybe they are better in some ways, but I don't believe that it is an industry first.
Pete Baldwin 12/5/2012 | 3:25:41 PM
re: Intel Detects More Silicon Photonics litsem -- You're mostly right; Luxtera's got silicon APDs in the lab. They tell me they haven't commercialized them yet. (See the first comment in the thread.)

The "first" that Intel is touting is the performance that exceeds that of silicon.

Luxtera, though, says they get performance that's better than Intel's -- although that's by a different metric, in terms of sensitivity (1dB better than Intel, they claim.)

Not sure Luxtera's in any rush to commericalize the silicon APD. They're really focused on short reach connections right now, and the APD just isn't that useful to them there.
Pete Baldwin 12/5/2012 | 3:25:41 PM
re: Intel Detects More Silicon Photonics You may be right; it looks like Eudyna's got one. My mistake.

I was talking more about commercially, currently available devices, but yeah, there have been papers about 10G APDs and probably some startup work in the early '00s that's dead.
bw 12/5/2012 | 3:25:40 PM
re: Intel Detects More Silicon Photonics Craig: anybody in the industry should know that 10G APDs had been in commercial use since late 1990s supplied by Agere (now CyOpitcs), Fujitsu (Eudyna), JDSU, etc. Just check every 80km 10G 300pin transponders: each has a 10G APD.
Pete Baldwin 12/5/2012 | 3:25:31 PM
re: Intel Detects More Silicon Photonics bw - thanks for the info.

Folks I'd talked to couldn't point to any 10G APDs, so obviously I needed to check around more (or maybe I misunderstood something.)
^Eagle^ 12/5/2012 | 3:25:29 PM
re: Intel Detects More Silicon Photonics Craig,

indeed, apd's at 10G have been around for quite sometime. Also there are APD's at 40G as well.

only thing new here is doing it in silicon (quasi silicon). What they don't tell you is that you actually cannot take advantage of a standard CMOS line and integrate silicon (active) photonics so easily as the Germanium they use to get the APD working is not used on standard CMOS line in this way. LOTS of additional tooling designs, fabrication design work and fundamentals need to be solved to make it work, and those special things are not currently used on CMOS lines. So, HUUUGE investment is still required to retool. Therefore, one wonders if indeed this will be a lower cost approach (silicon photonics.. CMOS but not really...) over traditional GaS and InP active photonics considering the EPI cost for most FABs currently in the world have been written off already due to the last 3 "downturns". Perhaps in the long term CMOS photonics will fly, but for now, the advantages are purely theoretical. Long way from changing the world. Look for lower cost InP approaches and hybrid approaches to dominate for quite awhile I think.

Sign In