Re-using embedded software has been the subject of
exuberant marketing for nearly two decades—and so far
it has amounted to little more than that. But in recent months,
quietly and far from the blaring hype, real-world testing is under
way to use software as a cost-effective means to embedding
functionality in a wide range of devices. Perhaps no where
is this development of re-useable embedded software more
evident than with the major players of the Power Architecture
(Power.org) community – especially with IBM.

The most experiment in re-useable software comes from a
joint effort between Samsung and IBM’s Haifa Research
Lab in Israel. If all goes as planned, IBM plans to begin
commercializing what is essentially a middleware framework
that allows embedded software to plug in and differentiate
semiconductors. That means in many cases it will no longer be
necessary to build new chips for every new device.

Exactly where this plays in the real world is unknown, but the
first beneficiaries are likely to be IBM’s partners in its growing
ecosystem, most notably the members of Power.org, which
includes companies such as Freescale, Sony, Cadence Design
Systems, Synopsys and Applied Microcircuits (AMCC).

The re-usable code project is based on what IBM calls the
Consumer Electronics Development Environment, known
inside IBM as COMPETENCE. At the heart of this
environment are architecture description languages, a class of
descriptive languages that includes Darwin, C2 and Acme, to
provide a high-level of abstraction for writing software. None
actually exists for the consumer electronics world, a fact that
IBM wants to change and to capitalize on.

Normally, embedded software is written for specially
developed processors with very specific functionality. With
COMPETENCE, a single chip design theoretically will be
sufficient to replace many designs, because the embedded
software modules will add different functionality.

The advantages are immediately obvious to anyone who has
designed chips for such areas as cell phones or MP3 players.
Each chip can cost many millions of dollars to develop, with
those costs escalating at each new process node. At 65nm and
45nm, non-recurring engineering costs render it impossible to earn a profit without an enormous sales volume. Adding
more functions only increases that cost, and it can slow the
time it takes to bring products to market, in large part because
of the difficulty of verifying the chip.

“The world of modeling is well known in hardware because
it’s so difficult to work with,” said Alan Hartman, research
scientist manager at IBM’s Haifa Research Lab. “In software,
this is new. But when you look at the cost of a luxury car, a
large part of that cost is the software.”

History repeats …sort of

In the realm of software, component-based systems engineering
is a relatively new field. Some of the groundbreaking research
was performed by Philips Electronics a decade ago using
various component models, and in the mid-1990s many
software companies including IBM had plans for software
objects—portions of applications rather than applications
themselves—that could be bundled together like Legos and
plugged into a framework.

The problem was that at the application level, these software
objects never played together like a single-vendor plugand-
play system. IBM, however, was large enough and had
enough of an investment in middleware that it could shift
the development to where there was a strong business need.
That need grew as the cost of developing ASICs and ASSPs
escalated into the tens of millions of dollars.

“If you look at multifunction printers, some have faxes,
scanners, and different speeds, and each is implemented by
a separate component,” said Hartman. “We can design a new
printer with new components so you only have to write a very
small amount of code.”

The approach plays off of the holistic design that IBM has
been pitching for the past several years. “What this allows
you to do is manage a product line as a whole,” said Julia
Rubin, an IBM research scientist for industry solutions in
the Haifa Lab. “You determine what features you want, and
only the pieces you want go to market.”

That works in principal, at least. But the real trick in this
scenario is what is relegated to software. Tony Massimini, chief of technology at Semico Research, said certain
functions are fine in software, particularly with a powerful
PowerPC processor running those functions. But he said
the model runs into trouble when it comes to video because
the response time has to be so high that it needs to be hardwired
into the processor.

“The PowerPC has a lot of performance, and if it’s not a
portable application you may get away with it in software,”
Massimini said. “If you look at a car, the response time versus
streaming video is easy to accomplish in software. With a
portable application, you can overload the processor.”

What ’s in the package?

What IBM and Samsung are developing goes beyond just writing
embedded software, though. Hartman said the companies also
are providing an editing environment, the ability to validate the
software with rules and to configure a particular product.

“Then, at the touch of a wand, you get general building
scripts,” he said. “Companies in consumer electronics have
already moved to a component-based architecture, so this
is a natural fit. What we’re also doing is taking legacy code
and componentizing it.”

When exactly that will become productized remains to
be seen, however. At this point, the legacy code work is
relegated to some future date. IBM will offer that type of
work as a service through its consulting group. However,
IBM researchers contend that fixing the bugs in software
is a relatively simple and inexpensive process, versus
trying to fix them in hardware. In part, that is because the
programming languages for software are relatively flexible.
C++, for example, is highly modular, compared with trying
to scour through massive amounts of verification data to
pinpoint a bug and then fix it in hardware.

Language isues

One of the reasons that ADLs remain relatively unknown in design
is that there is no single standard. As such, there is no agreement
on what needs to be included in the overall description.

IBM is looking to establish such as standard for the consumer
electronics industry, along with a modeling framework that
third-party components can plug into to provide consistent
behavior. IBM said it also plans to extend that framework
to other industries in the future.

At least part of that will be based on a standard modeling
framework, called the Unified Modeling Language. IBM’s
COMPETENCE relies on the Rational toolset (see
diagram), which allows designers to develop components
based on established models. But how quickly all of this
technology comes to fruition, and just how widely it is
deployed, remains to be seen. Building better software,
particularly for the embedded world, is not a new problem.

Real-world aplications

What this means for chipmakers such as Freescale,
interconnet makers such as AMCC, and large OEMs such
as Sony depends on when and how this technology hits
the market. Nevertheless, such moves could make both
companies more competitive because re-usable code means
faster time to market and lower NRE costs.

In Freescale’s case, for example, such code could be used across
a raft of vertical market applications, including automotive,
consumer electronics, industrial, and wireless, while for AMCC
the code would be used in embedded interconnect applications.
In Sony’s case, re-usable code could help cut NRE costs in the
consumer electronics market, where price is a key differentiator
for many of its competitors.