timing error detec on an adap ve scheme to combat

TimingErrorDetec.on:AnAdap.ve SchemetoCombatVariability - PowerPoint PPT Presentation

TimingErrorDetec.on:AnAdap.ve SchemetoCombatVariability EE241FinalReport NathanNarevskyandRichardOE Mo.va.on Asprocessnodes shrink,thereare


  1. Timing
Error
Detec.on:
An
Adap.ve
 Scheme
to
Combat
Variability
 EE
241
Final
Report
 Nathan
Narevsky
and
Richard
OE


  2. Mo.va.on
 • As
process
nodes
 shrink,
there
are
 .ghter
constraints
due
 to
process
varia.ons
 • What
are
the
 appropriate
ways
of
 comparing
the
 different
TED
circuits
 available?
 


  3. Razor
Latch
 • Main
latch
and
shadow
 latch
use
opposite
 phases
of
the
clock,
can
 check
to
see
if
a
 transi.on
occurs
that
 borrows
.me,
which
 means
the
path
driving
 the
latch
does
not
meet
 .ming


  4. Timber
Latch
 • When
enabled,
uses
the
 path
controlled
by
L
 • Creates
a
delayed
 window
to
allow
.me
 borrowing
to
correct
for
 errors


  5. Razor
FF
 • Detects
transi.ons
 during
the
.me
when
 CK
and
nCK
are
both
 high
using
the
dynamic
 or
gate.


  6. Timber
FF
 • Delays
are
controlled
to
 determine
a
specific
 amount
of
.me
 borrowing
and
error
 detec.on


  7. Setup
for
Analysis
 • Detec.on
window
–
Sweep
the
edge
of
the
 data
star.ng
from
right
before
the
rising
edge
 of
the
clock
into
the
clock
period,
enforcing
 errors
 • Repeat
over
a
range
of
supply
voltages
to
 determine
the
minimum
opera.ng
voltage
 • Measure
the
power
of
opera.on
for
circuits
at
 both
nominal
and
minimal
vdd


  8. Latch
Error
Detec.on
Width
Versus
 Supply
Voltage
 3
 2.5
 2
 Window
Width
(ns)
 1.5
 Timber
Latch
 Bubble
Razor
 1
 0.5
 0
 0.5
 0.6
 0.7
 0.8
 0.9
 1
 1.1
 Supply
Voltage
(V)


  9. Latch
Results
 • FOM
=
Twindow
/
 Timber Razor Latch 
 Latch 
 (P@Vmin)
 Clk‐>
Q
 300.6ps
 300.7ps
 delay
 • 5%
difference
in
error
 Power @1V 
 30.35uW 
 23.8uW 
 .me
window
 810mV 
 720mV 
 V min • 12.5%

Vmin
.mber
 19.1uW 
 14.1uW 
 Power @ V greater
than
Razor
 min 
 Nominal 
 2.31n 
 2.43n 
 • 35%
power
consump.on
 .mber
greater
than
razor
 FOM 
 1.209 
 1.723 


  10. Flip
Flop
Results
 • 75%
difference
in
error
 Timber FF 
 Razor FF 
 .me
window
 Clk
to
Q
delay
 300.6ps
 300.5ps
 Power @1V 
 32.03uW 
 43.9uW 
 • 4%

Vmin
.mber
 greater
than
Razor
 750mV 
 780mV 
 V min 17.63uW 
 26.5uW 
 Power @ V min 
 • 50%
power
 consump.on
.mber
 Nominal 
 37p 
 65p 
 greater
than
razor
 FOM 
 2.099 
 2.453 
 


  11. Conclusion
 • Razor
latch
out
performs
Timber
latch
for
all
 measured
metrics
 • Razor
Flip
Flop
allows
for
a
error
detec.on
 window,
but
uses
significantly
more
power
while
 also
opera.ng
at
a
higher
Vdd
 • Razor
latch
is
the
most
interes.ng
design,
and
 could
be
morphed
into
a
FF
with
a
hard
edge.

 • Both
FF
designs
have
their
advantages,
and
the
 use
of
these
TED
circuits
is
highly
dependent
on
 the
design
goals


Recommend


More recommend