Comms chips

Chipmakers Snack on Optical Designs

Companies that make semiconductor intellectual property -- chip design houses, essentially -- don't often become big names, but they've become awfully popular on the merger and acquisition circuit lately.

The buyers have mostly been field programmable gate array (FPGA) vendors -- companies who would often resell these startups' designs, or cores, to system vendors. Consider:

This work can be done by just a handful of people. TPack was big enough for AMCC to report a purchase price -- $32 million -- but the others were blips, as are many cores companies.

Why the sudden interest in these little outfits? Some of it has to do with chip economics. In optical networking, there's a need for new devices at 40 and 100 Gbit/s; Sarence, for instance, is a leading provider for a key interface called Interlaken. But the volumes aren't large enough to amortize the high costs of developing these chips on cutting-edge semiconductor manufacturing technologies, says Bob Wheeler, an analyst with The Linley Group .

On top of that, line cards often end up outfitted with an FPGA anyway, to connect to a proprietary fabric interface. The combination has led FPGA vendors to suggest that their chip absorb other line-card functions. "Clearly in the optical transport space, there's been a shift from ASSPs [application-specific standard products] to FPGAs," Wheeler says.

Some of it also has to do with Xilinx specifically. The company's Communications Business Unit, formed last year, is trying to sell devices that are more like ASSPs -- devices closer to a line-card-on-a-chip, in a sense -- and that generated a need to own some of these cores, says Gilles Garcia, the division's director of marketing.

Finally, these companies are simply there, a byproduct of the telecom crash. As chunks of big companies fell off into the sea, some semiconductor teams staying intact as startups. Modelware's founders were from Bell Labs ; Omiino's came out of Nortel Networks Ltd.

This doesn't mean there's going to be a rush to create new cores startups. "The problem is, you pretty much have to be acquired, because these companies aren't going to go public. But you can run a profitable little company," Wheeler says.

Not every cores startup has been snatched up yet. Wheeler thinks German company MorethanIP , which makes cores including 100Gbit/s Ethernet media access controllers (MACs), could be a target for Altera as a response to Xilinx's pickups. "Xilinx may have forced Altera's hand," he says.

There's also Sembarc Ltd. , an Irish company with an Optical Transport Network (OTN) bent, and Xelic Inc. , a cores company that's been working the optical networking market for several years.

— Craig Matsumoto, West Coast Editor, Light Reading

Pete Baldwin 12/5/2012 | 5:04:08 PM
re: Chipmakers Snack on Optical Designs

Neal - great to hear from you.  You might be right about ASICs suffering more than ASSPs.

I think it was also interesting to see AMCC buy TPack.  In a sense, these cores companies can be like distributed R&D for the chip industry.

I don't think that's a permanent trend, though. The circumstances around 100G just created the right circumstances.

Neal Neslusan 12/5/2012 | 5:04:08 PM
re: Chipmakers Snack on Optical Designs

While there may be an increase in the deployment of FPGA silicon in portions of the BOMS of optical telecom equipment,  I think a closer inspection will reveal tha FPGA's are winning not at the expense of commercially available ASSP's but rather at the expense of interally developed ASIC's (developed by the systems companies themselves).  I don't think the merchant telecom ASSP vendors,  especially those shipping volume stuff like 10G OTN chips,  are suffering at the hands of Altera and Xilinx.  Rather the ASIC starts are just not happening for the functions that are not yet widely available in the ASSP space (like 100G OTN) due to the high cost of the NRE's, so the result is the customer base resorts to FPGA designs.  The next two years will determine if the FPGA vendors can reverse the historical trend of the migration of a emerging technology (100G in this case) from FPGA/ASIC to ASSP as the industry matures.

Pete Baldwin 12/5/2012 | 5:04:07 PM
re: Chipmakers Snack on Optical Designs

What was in the back of my mind when writing this was whether Tabula might just swoop down and take a lot of business from all these guys.

It's a longshot, i know, in the same sense that any programmable-logic startup is a longshot. But if the intellectual property is that complex, and if ASIC costs are running that high... well, i know we've had this discussion before, but it seems like there might be a confluence of factors that would give Tabula an opening here, or maybe in the 400G/Terabit generation.

Neal Neslusan 12/5/2012 | 5:04:04 PM
re: Chipmakers Snack on Optical Designs

At the end of the day,  when you look at the life cycle of a typical telecom semiconductor "function" (for instance a framer),  the historical evidence suggests that at the early part of it's life it tends to be vertically integrated by the systems company,  be it by ASIC or by FPGA.  Later in time however,  and the past 10 years have greatly proven this out,  it tends to migrate to an ASSP (for a variety of reasons the least of which is not that the systems company can no longer afford to employ all those who have designed the original FPGA/ASIC).  Eventually it has moved to an ASSP and I would argue it will always do that because the cost of the FPGA will never make business sense for the BOM cost of the system if the vendor wants to reach "high volume" (note the quotation fingers).  I don't disagree with your assertion that the IP companies have facilitated the realization of these complex devices through the FPGA companies but there has yet been any market proof to show that the FPGA has staying power from a cost perspective.  If that were true then AMCC,  Cortina,  and PMC would not be shipping craploads of 10G OTN chips.  Is craploads a technical term?

I guess what I'm saying is that while I'm shipping a few hundred boards of 100G over the next three years it's perfectly fine for me to deploy my big massive Xilinx or Altera FPGA that costs me $4000 a whack,  dissipates 20-30 watts and is in a 45+mm package...  but my boss is breaking my ass to get to a lower cost,  lower power,  and smaller footprint when I get to thousands of boards per year and I don't know if the FPGA companies can provide that.

Sorry Gilles, my friend,  I only type what I truly feel! (smiley face)


Sign In