MICROELECTRONIC SYSTEMS NEWS

FILENUMBER: 1181 BEGIN_KEYWORDS Reconfigurable Systems END_KEYWORDS DATE: September 2004 TITLE: Reconfigurable Systems Will Emerge
=================================================================

TITLE: Reconfigurable Systems Will Emerge

Mobile Systems
Contributed by Nick Tredennick, Gilder Technology Report

Application emphasis is shifting from tethered systems to  mobile
systems. This shift changes the design goal from cost-performance
to cost-performance-per-watt. The microprocessor was well  suited
to  cost-performance applications, but it is not efficient enough
to be the  workhorse  in  mobile  systems.  The  answer  will  be
reconfigurable  systems,  with  the  microprocessor moving to the
role of supervisor controlling dynamically  configured  resources
that adapt efficiently to computational tasks.

For thirty years, the microprocessor has been  the  workhorse  in
digital systems. The microprocessor market has grown from nothing
to  billions  of  units  per  year.   But   the   general-purpose
microprocessor is reaching the end of its road. Voltage, which is
lowered to control power use, is approaching limits at which  the
microprocessor's  transistors  will  quit  working.  Further,  as
transistors shrink they become faster, but they also  leak  more.
Leakage  power is approaching active power for many chip designs.
That's important as applications shift to mobile systems  because
these systems typically want long standby times interspersed with
short periods of activity.  Many  reconfigurable  systems  failed
because  they  attempted  to change design methods. One legacy of
the  microprocessor's  success   is   that   problem-solving   is
programming.   Reconfigurable   systems  that  redefine  standard
programming languages or that require logic  design  skills  will
fail or will have only limited success.

Attempts to improve the microprocessor's  application  efficiency
led   to  design-time  configurable  microprocessors,  but  their
creation requires logic designers. The next step in this  process
is  run-time  configurable  microprocessors. Such microprocessors
offer   advantages   similar    to    design-time    configurable
microprocessors,  but  do  not  require  logic  designers.  These
microprocessors are generic  and  can  be  manufactured  in  high
volume at low cost and customized in the field.

Design-time and run-time configurable  microprocessors  are  more
efficient   than  general-purpose  microprocessors  because  they
replace performance-critical sections of a  program  with  custom
instructions  that  employ  application-specific execution units.
These configurable microprocessors speed only  critical  sections
of  the  program  and each custom implementation works for only a
narrow range of applications. But their success  will  offer  the
proof-of-concept  demonstration  that  reconfigurable systems can
deliver more efficient implementations.

Automotive Market
Contributed by John Watson, Element CXI

As the electronics  market  turns  from  industrial  to  consumer
markets,  many  companies believe one of the largest growth areas
is the automotive market where there is a captive  customer,  the
desire  for  built-in  functionality, a bundled sales process and
broad customer interest. However, counter  to  consumer  markets,
long  automobile design and test cycle times have made predicting
which    cellular     standards,     frequencies,     networking,
audio/video/3D/digital      radio/television     features     and
specifications  to  include  an  impossible  problem.   Yet   the
electronic industry builds only rigid,  unmodifiable embedded SOC
solutions with no ability to service or easily add improvements.

The solution starts by realizing that for most of the time,  most
automobile  electronics  sit  unused.  For  example,  the  seats,
windows and mirror electronic systems are seldom  used,  but  the
hardware  for  these  tasks is always available. The same is true
for other electronics in today's automobiles. Obviously,  if  the
hardware that isn't needed now but will be needed sometime in the
future could be removed until it is needed,  then  less  hardware
would  be  necessary.   A  new  breed  of  IC  is needed which is
capable of modifying its  internal  hardware  structures  to  map
itself   to   the problem  at hand and then change itself for the
next  application.   However,   this   modification   should   be
accomplished   not   by   running  many   software   instructions
through  an   ALU,   but   by   quickly  building  very  explicit
hardware.   The   increased  efficiency  of directly  creating  a
hardware  solution  can meet the much more  complex  requirements
of   wireless  communications,  infrared  warning  systems,   and
others,   but   still   keep  the  flexibility  provided  through
software.

Another part of the solution is to build this new kind of IC  out
of  "lego-like"  building  blocks;  so that if one building block
develops a defect, it can be replaced by another like  block  and
the  system  continues to work. This overlapping-in-time reuse of
hardware, and the utilization of  a  small  number  of  elemental
building  blocks  that  are interchangeable with each other means
future  automotive  systems  can  be  created   that   are   very
inexpensive,  easily adapted during the design, manufacture, test
and delivery of a  vehicle  to  provide  new  functionality,  and
extremely reliable.

2005 Conference on Reconfigurable Systems (ERSA)
Contributed by Toomas Plaks, London South Bank University

This conference, which will be held in Las Vegas during June  27-
30,  2005,  will address the topics discussed above.  The meeting
will  focus  on  the  different  approaches  in  engineering   of
reconfigurable  systems and implementing of algorithms, including
theory,   architecture,   algorithms,    design    systems    and
applications.

For additional information, access: ERSA-2005

=================================================================

Return to MSN Home Page

dbouldin@utk.edu