Some of the most innovative work in
science and technology today occurs at the boundaries between fields.
Although opportunities within certain mature fields may be diminishing,
the combination of such fields may be new, and can present a wealth of
possibilities for research and development. Similar things can happen
when different sub-fields or domains within a single field are
combined. This talk will argue that within circuits and systems,
several different domains which are normally kept separate can be
combined to advantage, producing new possibilities for research and,
hopefully, new practical benefits.
At the request of the organizers, the
talk will include some historical information about the beginning, 30
years ago, of what are today called “mixed-signal”
integrated circuits and systems. Although such systems are certainly
mixed-domain, they essentially consist of cleanly digital and cleanly
analog parts, joined by analog/digital interfaces. The talk will
suggest that other possibilities for mixing the analog and digital
domains do exist. Further, analog and digital are not the only two
domains that can be mixed within circuits and systems. It will be
argued that several other combinations may offer the potential for
fresh research. Several examples will be offered.
In one example, the domains of linear
and nonlinear operation are mixed to make possible circuits and systems
that are input-output linear, yet are internally nonlinear; another
example involves analog systems that are input-output time-invariant,
yet are time-varying internally, thus allowing external control of
internal properties without the output’s noticing. Both
approaches make possible a greatly enhanced dynamic range. A different
type of internally varying, externally invariant system makes possible
the dynamic variation of the internal system structure without causing
output disturbances. This can be taken advantage of in creating analog
systems with controllable power dissipation, so that always the minimum
power is being dissipated for the signal processing task at hand.
Another example involves the mixing of
the digital domain and the continuous time domain, resulting in
real-time digital signal processors which operate entirely without
sampling; thus, no signal aliasing occurs, and the quantization error
has spectral properties which are more desirable than in the classical
case. In contrast to the case with classical DSPs, activity in these
unclocked processors is inherently linked to activity in the input
signal. Thus, if the input activity is low or absent, the power
dissipation decreases accordingly.
A final example involves hybrid
computation in the context of VLSI and ULSI. Today’s technology
allows the implementation of a special-purpose analog computer with
significant capabilities on the same chip with a digital computer,
along with interfaces between the two; the result is a black box that
is as accurate as a digital computer, but is faster than the latter.
The slides for this talk, including
extensive references, will be available on the conference Web site.