April 26, 1999, Issue: 1157
Section: Technology Focus: DSPs
Prognosis: The Sky's The Limit -- DSP opportunities
abound, but will TI leave anything for others?
Darrell Dunn
While competitors in the DSP market differ over the question of how the
contest is shaping up and who among them will be left standing, one point seems
unassailable: As the decade winds down, digital signal processing is generating
some of the greatest industry fervor and innovation.
Semiconductor suppliers and analysts project that opportunities for DSPs in the
first decade of the new century may be unparalleled in electronics industry
history, with the possible exception of the microprocessor in the 1990s.
Shaping this view is the fact that upcoming generations of
personal-communications devices-from digital cell phones to Internet- and
GPS-capable PDAs-will need embedded processors capable of performing
digital processing in real time.
Whether the playing field for participants in the DSP market will become more
or less competitive will depend in part on the speed with which OEMs are able
to move to integrated system-on-a-chip solutions, which require multiple
processor disciplines.
The DSP market has grown at least twice as fast as the overall IC market in
the past 10 years or so, according to analyst Will Strauss of Forward Concepts
Co., Tempe, Ariz. "There's no doubt in my mind that trend will continue for at
least the next five to six years. That should capture the attention of any serious
semiconductor player," Strauss said.
From humble beginnings in the early 1980s as a technology that few
understood-and even fewer could envision as a major force in the
semiconductor industry-digital signal processing is now a must-have for any IC
supplier intent on playing in the emerging SOC arena.
While both the general-purpose and the function- and algorithm-specific IC
(FASIC) sections of the DSP market grew more slowly in 1998 than in past
years, they again outpaced the overall industry. The semiconductor market
recorded a 9% revenue contraction in 1998, while the market for
general-purpose DSPs grew 9%, to $3.5 billion, according to Forward Concepts.
The FASIC segment, which encompasses a wide range of non-programmable
fixed-function DSPs, grew 10% in 1998, to $5.1 billion.
Forward Concepts projects that the general-purpose market will grow at a
compound annual rate of 31% in the next five years, and the FASIC market will
grow at 30%, for a total of about $32 billion by 2003.
"The fact is that DSP is the most embedded of all embedded processors, and
consequently the DSP market is under-reported and understated," Strauss said.
"Revenue is also escalating in the area most often classified as micro-
peripherals. That segment totals in the hundreds of millions [of units] and
includes such DSP-function devices as modem chips, which are not classified as
DSPs."
But exactly what should be categorized as DSP revenue has become academic
to the many processor suppliers that have taken notice of the technology's
growing importance.
"There's not a single microprocessor that is not adding DSP capability," Strauss
said. "In addition, there must be 20 DSP IP cores being designed as we speak,
and only about a half-dozen of those have been announced. It's just
unbelievable."
Is TI's lead insurmountable?
Market leader Texas Instruments Inc. expanded its share of the
general-purpose programmable-DSP market to 47% in 1998, from 45% the year
before.
At the same time, competitors in the general-purpose-DSP arena, as well as in
ancillary DSP segments, have strengthened their efforts to slow down the TI
juggernaut and stonewall the Dallas-based company's efforts to create for itself
a semiconductor monopoly similar to what Intel Corp. has enjoyed with its x86
architecture.
TI has spent the past few years completing a metamorphosis from a
broad-based semiconductor supplier with other interests ranging from military
systems to notebook computers, to a company that is now centered squarely on
DSP and related analog components.
"This market opportunity continues to far exceed any of our wildest
expectations of 17 years ago," said Mike Hames, a TI vice president and
worldwide manager for DSP in Houston. "I certainly don't think in even our
wildest dreams we could have seen TI divesting itself of basically all its other
pieces to focus on DSP.
"I would like to tell you we had infinite wisdom," said Hames, who has been
involved in TI's DSP efforts since the 1982 introduction of the TMS32010, the
company's first programmable device. "We didn't, but we did know we were
onto something early on because without really going off and marketing [the
TMS32010], people were making their way to Texas to talk to us."
By 1994, with its annual DSP revenue at about $400 million, "a few anarchists"
within TI began to see the long-term potential, according to Hames, and talk of
a billion-dollar product line began to surface.
With sales of its general-purpose programmable DSPs likely to exceed $2 billion
this year, the DSP revolution has been completed at TI.
This revolution is now spreading to other shores.
The other three major participants in the general-purpose programmable-DSP
market recently formed alliances to combat TI's dominance. Also, as dozens of
microcontroller and microprocessor suppliers have incorporated DSP functions
into their devices in the past 18 months, numerous start-ups have been launched
in an attempt to find a crack in the TI wall.
But TI's position has become so strong that some observers believe competitors
will be hard- pressed to keep from losing additional ground.
"The fact is-and I've heard no credible argument against this statement-all our
competitors are weaker today than one year ago," said Thomas J. Engibous,
chairman, president, and chief executive of TI, at a recent analyst briefing. "The
classic competitors are regrouping. ... At the very least, we've quieted
competitors who used to call us the dinosaur of the industry."
TI has numerous advantages in the marketplace:
- Its existing architecture, led by its highest-volume family, the TMS320C5x, is
firmly ensconced in major markets, including digital cell phones, in which TI has
design wins with market leaders Nokia and Ericsson.
- As much as 80% of third-party DSP software and 70% of DSP development
tools operate with TI's DSPs.
- Two years ago, TI got a jump on competitors by launching its fifth-generation
TMS320C6x family, which employs a very long instruction word (VLIW)
architecture.
- In the past decade, TI has spent millions funding university DSP programs, and
in 1998 established a $100 million venture-capital fund for DSP entrepreneurs.
"We've consistently come under attack, but [they haven't had] a lot of success,"
Hames said. "When we started this 17 years ago, you could establish a new
architecture for an investment of a few million dollars. Today, the investment is
more than $100 million for not only the silicon but also support, software, and the
total effort to bring it to market.
"The bill keeps going up, and if you intend to be successful, you have to spend
the same amount of money whether you have 70% market share or 5% market
share," he said. "The ticket to entry has become pretty high."
Gamble, or a trump card?
In the most dramatic challenge to TI's dominance, the three major contenders
have moved during the past year to combine their efforts through development
alliances.
Lucent Technologies Inc., the second-largest DSP supplier, announced last June
that it will ally with Motorola Inc., the third-largest DSP supplier. Together, the
two begin to approach TI in terms of revenue for general-purpose DSPs, but
are not close in terms of numbers of customers or third-party support.
Lucent and Motorola have formed what is called the StarCore alliance, and last
November opened a joint design center in Atlanta. Earlier this month, the
alliance announced details of its first core product, the SC140, which is expected
to begin shipping in volume next year.
In February, Analog Devices Inc. teamed up with Intel to form an
as-yet-unnamed DSP alliance that will be headquartered in Austin, Texas.
Both alliances are aimed at creating fifth-generation DSP architectures to
compete with TI's TMS320C6x.
"They're looking for ways to effectively confront TI in the marketplace," said
Jeff Bier, general manager of Berkeley Design Technology Inc., Berkeley,
Calif. "By pooling resources and putting additional momentum behind a single
architecture, they do have a good chance of taking on TI more effectively than
in the past."
Despite some early consternation when TI revealed it would use a VLIW-style
architecture for its new family, the StarCore alliance has already announced
that it will use a similar instruction set. Analog Devices is using VLIW in its
high-performance TigerSharc, announced late last year, and "it certainly
wouldn't be surprising" if the ADI-Intel alliance adopts the technology,
according to Bier.
"There are a number of good technical reasons why VLIW makes sense, and it
has driven a bandwagon situation," he said. "VLIW offers more potential than
could be obtained by tweaking old-style conventional architectures, and offers
the promise of better compilers that are more readily able to use high-level
languages."
The ease of use and efficiency of new compilers has become a hot topic as
designers are increasingly demanding the ability to move from assembly
language to a C-compiler emphasis, Forward Concepts' Strauss said.
"There's not a single company that hasn't said their compiler is more efficient
than brand X," he said. "It will be at least another year before we can really tell
how good the compilers operate and how bug-free they are. You can do
simulations of chips, but simulation of compilers without chips is awfully tough to
do."
The alliance's first fruit
StarCore will begin shipping samples of the SC140 in the fourth quarter. The
300-MHz core integrates four multiply-accumulate (MAC) units, four arithmetic
logic units, and four bit field units.
The core provides 1,200 million MAC operations per second, or 3,000-mips
RISC operation, placing it in the same performance category as TI's
TMS320C6x, which provides about 500 MAC operations/s and 2,000-mips
RISC operation. The SC140 is in the process of being upgraded.
Lucent and Motorola plan to use the SC140 core to create both standard
devices and SOC solutions, primarily for the communications industry.
The StarCore alliance is also expected to aid the two companies' efforts to
expand their third-party tools and software support. Lucent and Motorola have
sold DSPs to a relatively small number of large customers-thus limiting their
third-party support.
"What we're enabling through StarCore is the first true dual source in the [DSP]
industry," said Thomas Brooks, marketing director for StarCore. "With two of
the top three vendors in the market working together, it's going to attract
interest, ... resulting in more applications in our libraries."
The combination of Lucent and Motorola's market share will also provide the
economy of scale needed to ensure success, Brooks said. "I think it's now a
two-horse race, Lucent and Motorola on one side and TI on the other," he said.
"With the StarCore alliance you have two DSP experts and two
communications experts. One plus one equals three. Intel is not bringing any
particular DSP knowledge to their party. That alliance [with ADI] is a zero plus
one equals one."
Market realities will play out differently, according to Robert DeRobertis,
manager of the 16-bit-DSP product line at Analog Devices in Norwood,
Mass."This is a two-horse race [between TI and ADI], and we're telling the
mass market that Analog Devices has a depth of portfolio that will not be
overmatched," he said.
ADI, like TI, serves tens of thousands of customers, and, also like TI, has
significant support from third-party software and tools companies, he said.
After watching TI increase its lead during the past few years, Lucent, Motorola,
and ADI all came to the conclusion that allying was the fastest way-and
possibly the only way-to stay afloat in the DSP mainstream, Strauss said.
"Lucent people have told me they simply couldn't come up with the money to
outspend TI on new development," he said. "Analog Devices, if they want to be
anybody in the digital-wireless market, needed to move into the top tier. The
Nokias and Ericssons won't look at a company that can't produce tens of
millions of units per quarter, and one available company out there with that kind
of clout is Intel."
Intel Inside consumer products?
The decision by Intel to join with Analog Devices to actively pursue the DSP
market has added a wild card, and it will take several years to grade its
effectiveness and be able to fully determine Intel's intentions, analysts said.
For years, Intel was reluctant to overtly embrace DSP technology, although it
had covertly integrated DSP functionality in the past few years in the form of
multimedia extension instructions in its Pentium processor line.
Intel, which earned $6.1 billion on revenue of $26.3 billion last year, probably
could have acquired all of Analog Devices-whose 1998 net income was $119
million on revenue of $1.2 billion-"for pocket change" if the Santa Clara,
Calif.-based MPU giant had simply wanted to buy its way into the
general-purpose programmable-DSP market, an industry source said.
Exactly what Intel hopes to accomplish, and whether it's in the DSP market for
the long haul, is debatable. Most observers believe the company does not intend
to become a catalog DSP supplier, but instead is seeking to create a
high-performance DSP architecture that will enable it to penetrate expanding
markets such as third-generation cellular phones.
"Intel now realizes that the bulk of computer power, even for PCs, in the
2003-2004 time frame is going to be in the form of connectivity and multimedia,"
Strauss said. "I think the Internet convinced them. The numbers are in the
staggering millions-tens of millions of ports and other related digital appliances
that will be requiring DSP."
The ADI angle
To date, little has been revealed about the Analog Devices/Intel architecture.
The companies say they will use Analog Device's 16-bit ADSP21xx core as a
starting point to create a low-power, cost-competitive core by the second half of
2000.
Analog Devices moved earlier this month to expand its current DSP line to
forestall customer erosion in the next 18 months, before the first products from
the Intel alliance come to market.
The company introduced three members of its ADSP218x family with expanded
memory and performance; provided the first details of its higher-performance
but code-compatible ADSP-219x family; and added a version of its
remote-access server solution that enables 24 modem ports.
"When we announced the relationship with Intel, we heard [TI] saying we were
in trouble because we were not supporting the existing code base," Analog
Devices' DeRobertis said. "They didn't realize we had a three-step plan to
acquire tools companies to improve the tool chain, come out with this
code-compatible product with two times the existing performance, and then also
provide a new architecture to open our DSPs to new market segments."
Can anyone else survive?
Because DSP technology has emerged as a must-have component in the
libraries of virtually all processor vendors, several smaller companies are
attempting to push their way into the fray.
Robert Conrad, DSP product line director at Analog Devices, believes the new
DSP entrants may have already lost their window of opportunity.
"I think those guys have already come and gone," Conrad said. "We don't see
any big customer announcements or momentum growing for their strategies. A
year ago, I was worried about guys with lots of money and good technology
want-ing to get into the space."
Kevin Stone, product marketing engineer at ZSP Corp., Santa Clara, said the
top-tier DSP players may still be in for a surprise.
"I'm glad they've counted us out," Stone said. "When we get the design wins and
stick them under their noses, they'll wake up to the mistake they've made. As
any new player in the DSP market, we face significant challenges, and it may
have taken us slightly longer than initially expected to get into some significant
accounts."
ZSP's first offering, the ZSP16401, which uses a dual-MAC structure to provide
about 200-mips performance, has only recently begun shipping in volume. The
ZSP16402, which provides 400-mips performance, is expected to move into
production next month.
More entrants
Other companies attempting to enter the DSP arena include BOPS Inc., which
uses an "indirect" VLIW instruction set for its core offering that combines as
many as 16 processing engines. Another player is Siemens Microelectronics
Inc., with its 32-bit TriCore architecture and 16-bit Carmel core.
In the microcontroller segment, companies that have introduced devices
combining RISC and DSP cores, or adding DSP functional blocks to RISC
devices, include Hitachi, Hyperstone Electronics, IBM Microelectronics,
Integrated Device Technology, Intel, LSI Logic, Motorola, NEC, and Quantum
Effect Design.
"It's what our customers have requested from us," said Peter Carbone, SuperH
product marketing manager at Hitachi Semiconductor (America) Inc., San Jose.
"Our customers needed further cost reduction, integration, reduction in power,
and board-space reduction, so we came up with a new architecture that was a
unified RISC/DSP architecture," Carbone said.
Hitachi's SH-2 DSP architecture, introduced in 1997, provides up to 66-MHz
performance. Last year, the company added the SH-3 DSP, which provides up
to 133-MHz performance.
Although Hitachi has made no official announcement, the SH-3 DSP family is
expected to be extended beyond 200 MHz, and it appears likely the company
will maintain its dual-architecture concept for the SH-4 family, which will debut
in the next year or so.
The single-engine SH-DSP architecture provides the best of the RISC and DSP
worlds while simplifying design and implementation, according to Carbone.
"You could use separate [RISC and DSP] processors, but interprocessor
communication is really a headache," he said. "Two cores also require two
different memory subsystems, which is expensive, and it's not an optimum
implementation in terms of cost, power, and board space."
The combination of RISC and DSP in SOC implementations will escalate in the
next five years, analysts said, as evidenced by TI's move in February to license
the 32-bit Jade and 64-bit Opal cores from MIPS Technologies Inc.
The company will marry the two cores with its own DSPs for use in consumer
appliances, such as set-top boxes, and in voice-over-IP communications
products such as switches and routers, according to Richard Kerslake,
worldwide system-level integration marketing manager at TI.
In addition to the MIPS processors, TI holds licenses to the 32-bit ARM cores
developed by ARM Ltd. and ARC Cores Ltd., and to the SPARC architecture
designed by Sun Microsystems Inc.
Copyright ® 1999 CMP Media Inc.
|
