|
by Tom Cantrell
Start ı Big
Stack Attack ı IP Silicon ı i-Who-Chip?
ı And the Winner isı ı Sources
and PDF
Reviewing the pile of information I brought
home from the recent Hot Chips 13 conference, I was intrigued by the
presentation "Low-Power Silicon Architectures for Wireless Systems"
by Jan Rabaey from the UC Berkeley Wireless Research Center. [1] Naturally
given the title, much of the presentation dealt with the arcane world
of wireless technology and the alphabet soup of standards that litter
the planet. However, it also had quite a bit to say about an issue
relevant to practically every system designer: should I do it in hardware
or software?
There was a time when the answer was
black and white. Software was more flexible than hardware but slower
and less efficient. Most problems were easily categorized as to the
proper solution (i.e., one alternative was feasible, the other ludicrous).
These days, things arenıt nearly as simple.
Blame technology for blurring the lines with, for example, ultrafast
processors and programmable logic. Blame competition for upping the
ante associated with making the best choice. The danger is summed
up by the old saw, "If all you have is a hammer, everything looks
like a nail." The problem, if you will, lies in the fact that
itıs now possible to approach almost any task with either a hardware
or software hammer.
Now, this is by no means a new phenomenon.
For instance, what designer isnıt familiar with the concept of a soft-UART?
It was many years ago that this formerly hardware-only I/O function
succumbed to bit-banging in software. Itıs the same for other simple
functions such as multiplexing a seven-segment LED or scanning a keypad.
Iım not just talking about digital functions,
either. The emergence of the DSP was a huge win for software aficionados,
replacing op-amps and banks of resistors and capacitors with tight
loops.
Fast forwarding to the present, it seems
thereıs practically no hardware target some clever programmer with
a hot processor canıt take a shot at. In the context of wireless communications,
some people consider software radio the Holy Grail. All you need are
a 2-GHz-plus ADC and DAC, a half-dozen BOPs worth of compute power,
and a few megs of code. There is a bit of a problem, though. Can you
imagine running something like that on a battery? I can, for a few
minutes anyway.
Increasingly, power consumption is a
key measure right up there along with price, performance, flexibility,
and all the rest. And, not just for battery-powered applications.
At the high end youıll find processors with built-in thermal monitoring
that self govern their activity to prevent meltdown. Nothing like
a few NOPs injected into the pipeline from time to time to cool a
processorıs feverish brow.
Rabaey made his point with a whimsical
example. Projecting current trends, he found that in 180 years all
our wonderful computing gadgets will consume the total energy in our
galaxy. And, dedicated hardware designs are much more energy efficient
than general-purpose computing machines (see Figure 1). Stick that
in your cup of Java and drink it.
 |
| Figure 1ı"When it comes
to saving power, dedicated hardware is the call. But, flexibility
demands a softer solution. The secret is choosing the mix that
optimally meets the demands of your application." [1] |
NEXT
Circuit Cellar
provides up-to-date information for engineers. Visit www.circuitcellar.com
for more information and additional articles.
For subscription information, call (860) 875-2199, subscribe@circuitcellar.com
or subscribe
online. ıCircuit Cellar, the Magazine for Computer Applications.
Posted with permission.
|