Modern development tools remove complications-but at what price?
To say that embedded development has changed in the past few years should rank as one
of the all-time great understatements. However, I would assert that the average
developer's lot hasn't changed nearly as much as certain other magazines would have you
believe. According to these publications, everyone has on their desk a 135-MHz Pentium
with 64M bytes of DRAM and a 4G-byte hard disk loaded up with $20,000 worth of CAD/CAE
stuff needed to do your job. This toolset, in turn, allows you to design an entire project
in ASICs and guarantees that it'll work right the first time-all without getting your
hands dirty or even working up a sweat. On the hardware side, the job is likewise a snap.
Nowadays there's a chip for everything, so all you have to do is plop the right one on a
board and-voilá-you're done.
Unfortunately, this engineering utopia begins to fall apart when it encounters the
reality of what working designers must deal with every day. Therefore, one job as a
designer is to tune out the rhetoric and play it smart as to where you spend that precious
capital budget. In this column I hope to help you in this endeavor by highlighting a few
problems with new technology you won't read about in the glossy sales brochures.
The surface-mount scam
To begin this quest, consider the lowly soldering iron. It wasn't long ago that you
could do productive work with a unit costing $10. Then, with the introduction of
thermostatically controlled irons, the entry price rose to about $100. Now, with
infinitesimally small surface-mount parts, the new entry price for a "soldering
system" exceeds $5000!
A couple of years ago, one of my engineers convinced me to spend $2000 on a
surface-mount rework station so we could do our own prototype work. Although the station
was serviceable, it wasn't long before a new part came out sporting 208 legs on 0.5-mm
spacing. Now, during prototype, my firm must pay an outside vendor $50 each time we need
to either install or remove one of these chips. The only alternative is to spend an
additional $7000 on a new soldering system.
Adding insult to injury, even when the professionals solder down these high-density
surface-mount parts, we occasionally have problems. Invariably a leg or two doesn't bond
properly. To detect this problem, I've developed what I call the X-Acto test, and it
always finds unsoldered pins. To perform this test you need an X-Acto handle with a sharp
#11 blade in it. Holding the flat side of the blade at 45° to the pins on a chip, slowly
and gently drag the knife across the pins while listening to the sound it makes. When the
blade hits an unsoldered pin, the normal tink sound becomes a thunk. The only problem with
this test is that you must be careful not to break the wispy legs off the chip.
Beyond such basic problems, surface-mount devices are also susceptible to other bizarre
problems. For example, a few years ago I received a shipment of 144 pin parts with a label
attached to the package warning us to bake the chips before soldering them down. Unable to
comprehend the reason for such a strange admonition, I called the factory and asked about
the label. The response was that the plastic in the IC body absorbs moisture. Hence if you
attempt to solder these chips in after storing them in a high-humidity environment such as
Maryland, Florida or the South Pacific, the parts could explode like popcorn. The baking
stage was required to slowly dry out the chips-though I still wonder what happens when
these chips start absorbing moisture while operating.
Finally, although surface-mount parts really look great and can cram a mind-boggling
amount of electronics into a small space, if you don't have high production volumes, the
assembly costs for soldering them down can easily exceed the cost of the parts themselves.
Illustrating this point, consider a small handheld meter we designed. Our contract called
for us to build 20 units, so as an experiment we built the board using surface-mount
devices. For about a dozen parts, the labor costs involved in building the board ran
approximately $33 each compared to an estimated $12 for a through-hole board.
ASICs for Everyman?
Another modern technology you often find in close proximity to surface-mount chips is
the ASIC. Recently I was reading an article in another magazine in which the author was
marveling at the fact that discrete FIFOs are still available. Given the relative ease
with which you can include a FIFO in an ASIC design, the author seemed amazed that these
components still survive. What this industry observer seems to have missed is that for
applications that don't run into the thousands of units per year and where new changes and
features arrive every month, ASICs are a neat idea that's as unreachable as the moon. For
small manufacturers the software for creating these devices is prohibitively expensive,
and the inflexibility of a chip designed to fill one specific niche doesn't fit in with
the realities of their market.
The one bright spot (well, sort of) in this picture is FPGAs, those ASICs programmable
at your facility. These neat parts allow you to squeeze an entire logic board within one
chip. However, development-tool costs are still a major problem. By the time you load up
all the software for logic entry, placement, routing and simulation, costs can easily
exceed $5000/seat-and you haven't yet bought the new hyper-fast PCs you need to run the
programs. To make matters worse, the hemorrhaging of your pocketbook doesn't end with
simply buying the product. You often must upgrade every year for an additional couple of
thousand dollars or risk losing support on the programs because they're now obsolete
Around my company, though, we don't have those funds lying around, yet we still must
design with these modern parts. Hence we continuously use "old" software that's
buggy and doesn't support the latest chips, but we're still able to design with it.
However, as new FPGAs come out, we've got to wait to use them until we can afford to
upgrade. To see how the industry got into this position, cast your mind back to the old
days when electronics companies sold electronics and software companies sold software.
Although this situation worked well, the lines soon began to blur. MMI, a device vendor,
gave away PALASM to make it easier for engineers to design with PALs and thereby sell more
chips. The idea for attracting engineers was simple-build the tools and they shall come.
Nowadays, try to get Xilinx or Altera to give away any software. Fat chance-as they see
the situation, their product is neither electronics nor software, it's an "integrated
solution." Unfortunately if Xilinx makes the best part for one portion of a design
but a different section cries out for an Altera component, you can quickly find yourself
in the oxymoronic position of buying two integrated solutions. Hence, today you can't
always use the best part for the job-instead it's now the best part you can afford the
Just plug it in?
A third area where problems await an unwary designer is with the crop of new chips that
implement such high-level functionality as voice synthesis or video display in a single
off-the-shelf device. No glue logic, resistors, capacitors, connectors, speaker amplifiers
or memories needed-just plug it in.
I once had a customer claim that he could build a complete system with just three
chips. One synthesizer chip would provide all the speech-generation functionality he
needed, a second part would implement the spread-spectrum transceiver the design spec
called for, while a third device used a complete PC to control the whole thing. It's hard
to tell a guy who's read all the ads that there's more to electronics than plugging
together ICs. There's also timing problems, power-up glitches, clock incompatibilities and
firmware bugs. Unfortunately, such misinformation is everywhere. Just look at the ads for
A/D converters, power-supply chips, flash memories or one-chip micros.
So, at the end of the day, how should we as engineers and designers respond to these
problem areas-and the dozens of others I haven't mentioned? Use the brain nestled between
your ears and avoid using a new technology just because it's there. Evaluate all the tools
that modern technology provides, but only use the ones that fit your application and