Design Boutique Helps Malleable
New York - A little-known network processor
start-up and a virtually unknown high-end processor design house that
have now come to light make up the true -not forecasted or predicted or
theorized-first day of the future of microelectronics.
Malleable Technologies Inc.'s recent Malleable Embedded Communications Accelerator (MECA)network processor
is one of the largest integrated ICs ever produced (the firm will not
say publicly exactly how large). It's a programmable logic- and
memory-packed system-on-a-chip (SOC) that came together by purchasing
third-party semiconductor intellectual property (IP) and outsourcing
millionaire design engineers, and then using an independent
billion-dollar foundry, Taiwan Semiconductor Manufacturing Co. (TSMC),
to fab it.
and the design teams did not use synthesis-the nearly indispensable
design automation step that turns hardware design language code into
structures-for large chunks of the design. And in another portentous
shift, the design engineers from Micro Magic Inc., Sunnyvale, Calif.,
who did this memory compiler work are mostly Ph.Ds who spent nearly all
of the 1990s designing microprocessors, not networking ICs.
|Mark Santoro, Micro Magic Inc.|
Whitney, Malleable's vice president of engineering, a bit reluctantly
discussed Micro Magic's part in the genesis of MECA, which took place
over only the last 12 months. Malleable, based in San Jose, took
private and corporate funding last June, said Whitney, though it will
not name those investors.
The super-size MECA voice processors
are meant for voice over asynchronous transfer mode (VoATM) and voice
over Internet Protocol (VoIP) equipment. Each device can process up to
96 channels of compressed voice or 512 channels of uncompressed voice,
including ATM or IP packetization. The MECA architecture is said to
integrate DSP- and packet- processing functions and replaces more than
10 existing devices, including eight general-purpose DSP chips
typically used in current designs. It's known that the transistor count
is above 20 million-a space where Pentium IIIs are barely treading.
Beyond that, comparisons are tough.
Malleable's Whitney would
only say that the Sonics Inc. SiliconBackplane IP was licensed and that
Micro Magic completed the design in large part by virtue of the
MegaCell memory compiler technology it has. She did not want to discuss
whether the design was a DSP array, as many firms are proposing, or how
much memory it has on-chip. She said the part is "programmable, but
customers won't program it."
"It's a large chip, and it has a lot
of memory," she said. "We're trying to stay away from talking about
transistor counts. We're shying away from a technology announcement."
Whitney and Micro Magic said they did not want competitors to know too
much about the design, though it would take those same competitors
"months to catch up."
Whitney said the memory in question is
embedded SRAM, done by Malleable's foundry TSMC. Micro Magic is now
selling a MegaCell compiler that could design such SRAM for $150,000, a
comparable price for any other EDA tool on the market. Asked whether
the architecture approach is anything like the new league of multi-DSP
platforms from Improv Systems Inc., BOPS Inc., or Infinite Technology
Corp., Whitney declined comment.
"It's not a chip that
Malleable's competitors (who are also the legion of NPU players) are
going to have an easy time figuring out," said Mark Santoro, president,
chief executive officer, and founder of Micro Magic. "It's very
different from those other approaches." Santoro bills MMI as the crème
of the crop, where nearly all the designers are Ph.Ds from Stanford,
the Massachusetts Institute of Technology, and the like.
Magic for the first time worked the disclosure of its work into a
contract, Santoro said. Since the firm's founding in 1995, clients had
MMI under nondisclosure, he said. Santoro is now looking to
commercialize the specialized datapath generator and layout control EDA
tools his designers developed in order to do their high-end work, so
the "latest and greatest" splash that Malleable is making should help
"The MECA processor is just a really good part for us to go
public on," said Santoro. Micro Magic is now seeing a lot of network
processor start-ups-though these days more and more are wholly owned
subsidiaries-who need this kind of expertise, he added.
"We're seeing a flood of start-ups, now that the VCs (venture capitalists) have found out about us," he said.
only question now is will MECA work as the intended customers need it
to? Cary Ussery, president and CEO of Improv, Beverly, Mass., is not so
sure. "In some ways, this is taking general ideas for FPGAs and moving
it up a level. My general opinion is that, while the technology looks
good on paper for 'pure' algorithms, real-world applications have more
complex requirements," Ussery said. "Malleable's technology may work on
small algorithmic components, but it won't scale well to full-blown
products. I do think this technology can be put together as a possible
alternative to the general reconfigurable FPGA approaches being worked