Efficient Switching Window Computation for CrossTalk Noise - PowerPoint PPT Presentation

1 / 28
About This Presentation
Title:

Efficient Switching Window Computation for CrossTalk Noise

Description:

Iterative solution: overall run time of O(n2) Multiple solutions. n. nx. k. l. m. l. m. 1 ... Linear run time is preserved. Time roll back very rare ... – PowerPoint PPT presentation

Number of Views:54
Avg rating:3.0/5.0
Slides: 29
Provided by: tauwor
Category:

less

Transcript and Presenter's Notes

Title: Efficient Switching Window Computation for CrossTalk Noise


1
Efficient Switching Window Computation for
Cross-Talk Noise
  • Bhavana Thudi
  • David Blaauw
  • University of Michigan, Ann Arbor

2
Traditional Delay Noise Analysis
Aggressor 1
Cc1
K1Cc1
K2Cc2
Victim
Cc2
Aggressor 2
  • Delay Noise Increases/Decreases Delay
  • Switch factor based noise model
  • Iterative Timing Window Propagation

3
Outline
  • Introduction
  • Delay Model
  • Time Sort Algorithm
  • Results
  • Conclusion

4
Introduction
Motivation
1
1
n
l
l
nx
1
2Cc
2Cc
Cc
m
m
k
0Cc
0Cc
  • Iterative solution overall run time of O(n2)
  • Multiple solutions

5
Introduction
Motivation
1
3
n
l
l
nx
2
4
2Cc
2Cc
Cc
m
k
m
2Cc
2Cc
  • Iterative solution overall run time of O(n2)
  • Multiple solutions

6
Introduction
Motivation
1
n
l
l
nx
2Cc
2Cc
Cc
1
m
k
0Cc
m
0Cc
  • Iterative solution overall run time of O(n2)
  • Multiple solutions

7
Introduction
  • Issues with iterative Approaches
  • Overall run time of O(n2).
  • Multiple solutions
  • Switch factor based delay model
  • Proposed Approach
  • Run time of O(n).
  • No multiple solutions
  • Superposition based delay model

8
Introduction
  • Superposition based model
  • Linear driver models Driver gate modeled with
    Thevenin resistance and Ramped voltage source

9
Introduction
  • Superposition based model
  • Linear driver models Driver gate modeled with
    Thevenin resistance and Ramped voltage source

10
Introduction
  • Superposition based model
  • Linear driver models Driver gate modeled with
    Thevenin resistance and Ramped voltage source

11
Introduction
  • Delay noise is a function of timing windows at
    victim and aggressor node inputs
  • Removes cyclical dependency between victim and
    aggressor net.

Switch factor based model
Superposition based model
12
Introduction
  • Delay noise is a function of timing windows at
    victim and aggressor node inputs
  • Removes cyclical dependency between victim and
    aggressor net.

Switch factor based model
Superposition based model
13
Outline
  • Introduction
  • Delay Model
  • Time Sort Algorithm
  • Results
  • Conclusion

14
Delay Model
  • Event (t, s)
  • t event time time when waveform crosses
    switching voltage
  • s transition time

vsw,r Vt,n
s
Vdd
vsw,f Vdd Vt,p
0.8Vdd
Early event (te, se)
vsw,r
t
Late event (tl, sl)
0.2Vdd
Timing window (te, tl)
15
Computing Noisy waveforms
  • Late event Noise pulse placed at (vsw,r vn)
    point
  • Early event Noise pulse placed at (vsw,r - vn)
    point

Noiseless event time tvl
Noisy event time tvne
Noiseless event time tve
Noisy event time tvnl
vn
vsw,r
vsw,r
vn
16
Delay Model Properties
  • Property 1 tvi lt tv

Victim input (vi)
Victim (v)
Driver input time is earlier than output event
time
17
Delay Model Properties
  • Property 2 tai lt tn

n
l
Victim
k
m
Aggressor input
Aggressor
Aggressor input time earlier than the start of
the noise pulse
18
Delay Model Properties
  • Property 3 tai lt tvn

Noisy victim time later than the aggressor input
time
Aggressor input transition
Vdd
vsw,f
Victim noiseless transition
Victim noisy transition
vsw,r
tvn
tv
tai
19
Outline
  • Introduction
  • Delay Model
  • Time Sort Algorithm
  • Results
  • Conclusion

20
Time Sort Algorithm
  • Uses Linear Superposition and Time Sorting

n
l
t
t
t
Condition 1 Arrival time of new event being
scheduled tvn gt t Condition 2 When scheduling
new event vn, no existing event of this type has
already been processed
n
m
k
k
l
m
t
21
Victim Early Event
n
l
  • Case 1 taiopt lt t

k
m
tvie t
n
l
tv
tvp
tvn
k
taie
taiopt
22
Victim Early Event
n
l
  • Case 1 taiopt lt t

k
m
tvie t
n
l
tv
tvp
tvn
k
taie
tail
taiopt
23
Victim Early Event
n
l
  • Case 2 taiopt gt t

k
m
tvie t
n
l
tv
tvp
tvn
k
taie
tail
taiopt
24
Victim Early Event
n
l
  • Case 2 taiopt gt t

k
m
tvie t
n
l
tv
tvp
tcp
k
taie
taiopt
25
Properties
  • Time progresses forward
  • Time roll back may be required in two scenarios
  • Roll back for a net can occur only once
  • Linear run time is preserved
  • Time roll back very rare
  • Early and late edge of a node processed only once

26
Outline
  • Introduction
  • Delay Model
  • Time Sort Algorithm
  • Results
  • Conclusion

27
Results for ISCAS85 circuits
Iterative Approach
Proposed Time Sort Algorithm
28
Conclusion
  • Method to compute the switching windows in the
    presence of crosstalk
  • Run time linear with circuit size
  • Eliminates Multiple solution problem
  • Significant Speed up
Write a Comment
User Comments (0)
About PowerShow.com