Comms chips

Fujitsu Fuels Net Processor Hopes

Applied Micro Circuits Corp. (AMCC) (Nasdaq: AMCC) today announced a network-processor design win with Fujitsu Ltd. (KLS: FUJI.KL), extending a streak of design-win announcements for network processor companies (see Fujitsu Picks AMCC's Processors).

Neither Fujitsu nor AMCC would reveal which devices are being used in Fujitsu's GeoStream R900 line of IP routers, but it's some combination of AMCC's OC48 chips, which include a network processor, traffic manager, and switch fabric. Jeff Cashen, vice president of AMCC's switching and network processing division, would not discuss how much money the Fujitsu contract represents, although he asserts that it's one of AMCC's bigger deals. "It's up there, certainly," he says.

It's been rare for OEMs to admit to using network processors, but that could change starting this year. Just this week, Silicon Access Networks Inc. revealed a design win (see Silicon Access Nabs Huawei), and nearly every network-processor vendor claims to be on the verge of announcing a deal with a Tier 1 customer.

For the past two years, network processor vendors have been under a virtual gag order regarding design wins, because those Tier 1 customers supposedly wouldn't give permission to be named. Many were just beginning to switch from home-grown ASICs to off-the-shelf network processors and weren't comfortable admitting it; others considered their sources trade secrets.

And for some OEMs, that's still the case. "For a lot of guys, they took their crown jewels and they transferred them to us. They don't want anybody to know until they've released the box, until somebody can buy it, tear it apart, and see what's inside," Cashen says.

But OEMs are beginning to loosen up, as network processors become more of an industry fixture. "There's no reason now for the systems guy to hide what's inside the box," says John Metz, principal analyst with Metz International Ltd.

Another factor is that OEMs don't like to talk about systems until they've hit the market -- and that's only now happening for the first wave of network-processor boxes. For example, the Fujitsu deal was one of AMCC's earlier network-processor contracts, with discussions that began "a couple of years ago," Cashen says.

— Craig Matsumoto, Senior Editor, Light Reading
gea 12/5/2012 | 12:38:53 AM
re: Fujitsu Fuels Net Processor Hopes As far as I understand it, using a netowkr processor is kind of like using a Pentium chip in a PC...that doesn't tell you a lot about what's running on the computer.

Likewise, presumably company's like Fujitsu piece together all sorts of code (some bought some developed in house) to create any "value-added". Once the speed of such processors cross's a certain threshold, who gives a crap if a NP is being used?

So...why exactly have these vendors been shy about using NPs? ANy practical reasons?
xip42 12/5/2012 | 12:38:43 AM
re: Fujitsu Fuels Net Processor Hopes I think the NPs available right now are very limited in many ways. This means each have unique features and pitfalls that can be exploited or can hamper the product being designed.

If you are a company and you have studied all the available NPs, and you hear your competitor is using a particular device (or actually the company is enough info). You may be able to infer available feature set.

For example if i knew my competitor was using Company X's NP, i could know from my own experience studying Company X's NP, that they cannot support line rate for some applications. They can "value-add" software to their hearts delight, but do to some "feature" in the architecture of the NP, they can't do it.

So simply identifying a vendor or part really can provide a lot of information about the product you are designing.

With a Pentium this is not the case. Simply put the NPs are so radically different in architecture, programming methods, and features, that simply identifying your NP is very revealing.

metroex 12/5/2012 | 12:38:38 AM
re: Fujitsu Fuels Net Processor Hopes xip42's point "that simply identifying your NP is very revealing" is very true. It's so true that if NP becomes more popular, then there would be only a few "standard" line cards in the market for each major application(s). Moreover, OEM's product differentiation would retreat from line card technologies to network software and OSS. Futhermore, interesting than not, a new business may be evolved, that the OEMs may outsource those "standard" line cards to the third party design houses to take advantage of economy of scale (lower cost)and focused expertise (shorter TTM). My 2-cent. :-)

Sign In