EBN Home Page
Site Search:
EBN News
This Week In EBN
Daily News Digest
EBN Resources
Tech Center
Business Center
Purchasing Center
Distribution Center
Wall Street Watch
EBN Columnists
Calendar of Events
Interactive
Stock Lookup
Message Center
Bookstore
Feedback
Salary Calculator
EBN Online Info
Search/Archives
Subscribe
EBN Media Kit
The EBN Editors
EDTN
Network Links
Search EDTN
News Center
Technology
Business & Finance
Products
All Headlines
EDTN Design Center
EE Design Online
Ideas for Design
Technology Labs
Reference Library
Analog Avenue
Embedded Systems
PLD Center
Prototype Center
Career Center
Job Listings/Career
Info!
Electronic Design Job
Bank
Network Partners
Design & Reuse
Electronic Buyers'
News
Electronic Design
EE Times
Partminer
Power Designers
SBN
EDTN Direct
EDTN on Pointcast
Desktop
Pricing/Availability
EBN's Daily News Digest
TI maps out broad approach for DSPs
Electronic Buyers' News
(10/27/98, 04:38:39 PM EDT)
A DSP with peripheral microcontroller functions, or one chip with DSP and
microcontroller cores? Although Texas Instruments Inc.'s take on the
technology is DSP-centric, the company sees room for several options,
including stand-alone microcontrollers.
Rod Trautman, DSP digital-control-systems manager at the Dallas-based
company, explains how he sees the market evolving.
EBN: Can you accomplish what is needed in embedded-control applications
with one or the other technology, meaning DSP or MCU exclusively?
Trautman: Embedded control is pretty broad. There are some applications
in embedded control that are best handled with a DSP solution, but there are
some applications for which the microcontroller is best suited. And then there
are some applications that are going to, quite frankly, require both.
EBN: In which applications is DSP technology providing the most
cost-effective solution?
Trautman: Embedded-control applications are moving more and more
toward real-time control. An example of this would be in digital motor control.
As you provide more and more performance through a DSP, customers are
able to take advantage of that in things such as motor control and reduce the
overall system cost and increase the performance and capability of the
motors. A DSP is going to do it in real-time control. What a DSP does very
well is digital filtering, which boils down to math-intensive multiply-accumulate
[functions]. Well, those are also heavily found in real-time control algorithms.
So what we've done to direct some of the control phases with DSP is we've
got the DSP, and then we put around that some of the traditional control
peripherals such as timers, analog-to-digital converters, serial ports, those
types of things. And that makes an ideal control chip for many applications. In
the case of motor control, [our customers can] use less-expensive motors.
They're eliminating components such as sensors, because now they don't
need to have sensors to tell everybody what the motor is doing. They can
detect and predict it using the advanced calculations of the DSP.
EBN: When is a microcontroller the best solution?
Trautman: There are some embedded-control applications that will
always use microcontrollers. Something that needs a simple control algorithm,
where you're taking some input from a user and you're going to go turn things
on [and] turn things off-that type of thing. There are no 4-bit DSPs. There
are no 8-bit DSPs. So, 8-bit microcontrollers and 4-bit microcontrollers are
going to be around for a long time. I think they'll continue to decline, but there
are remote controls out there for the TV that are going to continue to use
those types of baseline controls. They're not doing anything that's
math-intensive, that's real-time control-based.
EBN: What about combining the two cores on one chip?
Trautman: There's a third space where the system is complex enough
that you really need both. In a digital wireless phone, there's a DSP that's
doing many of the voice-type capabilities, the heart of the phone, the control;
and there's also a microcontroller that's doing some of the
man-machine-interface-type aspects. In that space, we see both processors.
In the case of the microcontroller, TI's using ARM cores. How will things
migrate over time? Will DSPs win out? Will microcontrollers win out? Will it
be that everything needs to go to a micro and a DSP on a single chip? The
answer-and it's not a real straightforward answer-is that all three of those
things are going to happen. Microcontrollers are going to have a place in
certain applications. And DSPs are enabling a lot of new applications,
[providing] a higher form of control. And there are some applications that are
just going to require both, because there's just so much going on that, quite
frankly, a uniprocessor going to a single core is not going to solve the need.
EBN: What's the most important thing for the OEM customer to know about
the advantages and disadvantages of the different alternatives?
Trautman: They need to look at their system. What problems are they
trying to solve? And if one of the key things that they want to do is real-time
control, and they want to really use some of the high capability that TI gives
them in DSPs-the number of mips that we can provide-if they need that in
their system, then they need to have a DSP in there. What we find out in
many cases is, if the microcontroller-type functions are a very small part in
the overall system, there's plenty of DSP mips to simply not have a
microcontroller in that system. If they're on the other end of the spectrum,
where you have a lot of interrupt control, you've got to do a lot of processing
of interrupts, you've got a lot of man-machine interface, you've got a lot of
if-then-else-type statements to go run in your code, you're doing an interrupt
control base, then it's going to be a microcontroller function. [Designers are]
going to continue to use those 8-bit and 4-bit and, in some cases, 16-bit
microcontrollers. And even 32-bit RISC. If that's what they need to use.
They're not trying to do something real-time. And then what you see in some
of the higher-end systems, you actually need both.
EBN: What technological developments are essential to such high levels of
integration and performance?
Trautman: In late August, TI announced 0.07-micron CMOS technology
that we're currently developing. It's going to allow you to pack more than 400
million transistors on a single chip, with up to seven layers of metal. This is
something we're targeting for production in the year 2001. When you look at
that, you can really do some system-online integration for some of these
applications such as wireless. [This type of integration entails] higher- and
higher-performance DSPs and higher- and higher-performance
systems-on-a-chip. The first product [in this area will use] SPARC
technology from Sun, and then shortly behind that should be the [incorporation
of] high-level DSPs.
EBN: For this advanced technology that you're targeting in 2001, what kind
of cores would you expect to be on there?
Trautman: It depends on the application. In some applications, yes, you'll
have both [DSP and microcontroller cores]. In many applications, quite
frankly, there will be just DSP, a very high-performance DSP. If you look at
some of the higher-end processors and DSPs we've introduced, I mean the
6000 processor goes up to how many mips? 2,000 mips? For the TMS326000
family that we recently introduced, we have versions up to 2,000 mips. And
that's not even using this advanced process technology. That's just going to
continue to increase substantially as you drive the technology down to smaller
and smaller geometries and higher and higher performance. The process
technology will enable up to 1-GHz operating frequency. When you get things
running that fast, you can do a lot.
EBN: What can you do?
Trautman: You can do many, many channels of modems because you'll
have a very high performance. In the wireless area, you'll be able to integrate
not only your basic digital GSM-type phone features. [You can] essentially
integrate all the other personal communicator-you know,
handheld-PC-type-functions all into a phone.
EBN: Could you tell me a little bit more about the technology that's going to
lead to this advanced system integration-dubbed system online integration?
Trautman: As you shrink the geometry, you're going to be operating as
low as 1 V. As the voltage goes down, the power goes down. So that's going
to enable higher and higher performance, lower and lower power. The
[decreased] power is going to be a huge benefit to those applications that run
on batteries, such as cell phones. So those types of things are going to bring
lower cost, more performance, and lower power dissipation for all the
embedded-control products.
EBN: TI is a leader by far in DSP, and you have an established customer
base that's accustomed to working with TI on DSP. Is it going to be a
challenge to get customers up to speed on microcontroller tools as well if an
application requires both cores?
Trautman: We don't see all applications going toward a dual-core solution.
There's going to be a large portion of these things that are DSP-based; it's
going to be a mix of everything. The toolset that is used between, for
instance, the DSP and the ARM core is a common tool chain. So, the tools
that the customers use, such as the user interface-the graphical user
interface-and the debug environment are very similar. So it makes it very
easy for the customers when they do need to use both, in those few cases
where they're putting both of them together in the system.
EBN: Can we make price comparisons for the different DSP/microcontroller
options?
Trautman: It's real tough to do because you really have to look at the
application and what's required for the different applications. Because there's
so many factors that get involved: what level of integration have you done,
what peripherals have you put on, and how much memory have you put on it?
Have you put flash? Have you put RAM, ROM, whatever? Do you use
off-chip memory? A lot of people perceive that DSPs are more expensive
than microcontrollers. And, quite frankly, you've got to make sure you're
comparing apples to apples. Look at a 16-bit DSP that provides 30 mips of
performance and has peripherals on it and the price is $4, compared with a
16-bit microcontroller with similar peripherals that runs at 2 or 3 or 4 or 5
mips; and you sit there and say, well, that microcontroller is 50 cents less than
a DSP. But that DSP can do so much more [in terms of] the performance it
provides, the flexibility that that software will provide running so fast, and
overall system cost reduction. So, when you get into an apples-to-apples
comparison at the system level, the DSP is definitely at cost parity; in many
cases, it's actually going to reduce the overall system cost. Many of our
[OEM] customers are going to DSP for embedded control because they can
reduce the overall system cost while increasing performance and satisfying
the end customers. That's utopia-you get a lower cost and higher
performance and make everybody happy
EBN: How about the cost of combining the functions?
Trautman: Obviously, the more you put around the device, the more
costly it is. If you try to put that same functionality the other way-around an
MCU-it's going to be more costly than putting it around a DSP, because the
DSP starts out from a higher performance point. If you add something, it
takes up silicon area which translates into cost. The advantage that we offer
is that we believe that we're offering higher performance without adding a
bunch of functionality from the standpoint of silicon area, so we can provide
that same price yet give them more performance.
EBN: Is there anything else the customer should know in making the choice
of what type of device to use-to use a discrete DSP, a discrete
microcontroller, a DSP with some control functions, or a single chip with both
DSP and microcontroller cores?
Trautman: They need to look at the main criteria of their system [and
where it's] going to be in two years. For instance, in motor control, customers
are able to find out that they can eliminate sensors by using DSPs. You've got
to look beyond the silicon when you actually look at how you want to
architect your system. Not really beyond the silicon, but beyond the
controller-the DSP, the MCU-beyond that control function. Look at the total
system and how you want to do that, and that is the most important thing to
do. And look at your key care-abouts. And make sure you choose the product
family that's going to continue to provide you more and more performance
and further cost reductions as you continue to go down the product roadmaps.
EBN: What other critical changes are coming about?
Trautman: Currently, antilock brake systems have a high-performance
microcontroller in there. These systems will migrate to higher and higher
performance, but they're also going to migrate to where they're going to go to
brake-by-wire [technology]. They're going to replace the hydraulics in the car
and use electric motors to apply the brakes on each wheel. Many customers
will use TI DSPs to do that high-performance motor control. That type of
system will require so much performance [that] they'll use a DSP, which is
well suited in the control space for that embedded control. There'll probably
still be a microcontroller in the system that will do the overall control-based
algorithms-the overall system processing, so to speak-and the DSPs will do
the real-time controlling.
EBN: And the DSP will have some microcontroller functions on it?
Trautman: Microcontroller functions from the standpoint of peripherals.
Timers, a-to-ds, those types of things. But in that system, there'll still be a
microcontroller. In this case, a TI microcontroller. But there will be a central
microcontroller that will kind of be a system processor, and the DSPs can
focus on the real-time control-signal processing-high-performance control of
the electric motors, in this case. We see a lot of that occurring in many
applications. Wireless is a place where TI is already integrating DSP and
MCU together, and it's in production.
Editor's note: TI already has single-chip devices that combine DSP/MCU,
ASIC, logic, and memory functions. This technology is geared to digital
baseband applications. The 0.07-micron process will enable the incorporation
of other functions, such as analog and flash cores, according to Thomas
Brooks, TI's marketing manager for digital cellular applications.
Will this technology be ready in 2001 for high-performance baseband
applications? If the customer is, Brooks said.
|
