Combining Decomposition and Unfolding for STG Synthesis (application paper) Victor Khomenko 1 and Mark Schaefer 2 1 School of Computing Science, Newcastle.

Combining Decomposition and

Unfolding for STG Synthesis(application paper)

Victor Khomenko1 and Mark Schaefer2

1School of Computing Science,Newcastle University, UK

2Institute of Computer Science, University of Augsburg, Germany

2

Asynchronous circuits The traditional synchronous (clocked) designs

lack flexibility to cope with contemporarydesign technology challenges

Asynchronous circuits – no clocks: Low power consumption and EMI Tolerant of voltage, temperature and

manufacturing process variations Modularity – no problems with the clock skew

and related subtle issues[ITRS’05]: 22% of designs will be driven by ‘handshake

clocking’ in 2013, and 40% in 2020 Hard to synthesize efficient circuits The theory is not sufficiently developed Limited tool support

3

Syntax-directed translation

Idea:

Convert the specification to a network of standard handshake components (Balsa, Tangram)

Computationally efficient Solution is guaranteed Produces highly over-encoded circuits, with

large area and low performance

4

Logic synthesis

Idea:

Synthesize the circuit by exploring the state space of the specification

Produces good circuits Solution is not guaranteed State space explosion: synthesis

based on state graphs is feasible only for small specifications (20-30 signals for BDD-based Petrify)

5

Unfoldings

Alleviate the state space explosion problem More visual than state graphs Proven efficient for model checking Can often synthesize specifications with

100-200 signals Still not enough for real-life designs!

6

DecompositionIdea:

• Decompose the control path of the specification into smaller clusters and synthesize them one-by-one

• Use syntax-directed translation for clusters on which synthesis fails

Can halve the area of the control path and improve its latency [Carmona, Cortadella DAC’06]

7

Example: VME Bus Controller

lds-d- ldtack- ldtack+

dsrdtack+ d+

dtack- dsr+ lds+

DeviceVME Bus

Controller

lds

ldtack

d

Data Transceiver

Bus

dsrdtack

8

Initial partition


dsrdtack+ d+

dtack- dsr+ lds+Include signal triggers

and choices:• lds: dsr, ldtack, d• d: ldtack, dsr• dtack: d

lds: dsr, ldtack, d

d: ldtack, dsr dtack: d

lds

d dtack

dsr

ldtack

9

Initial decomposition


dsrdtack+ d+

dtack- dsr+ lds+


dsrdtack+ d+

dtack- dsr+ lds+


dsrdtack+ d+

dtack- dsr+ lds+

10

Irreducible CSC conflict

Transition contraction


dsr- d+

dsr+ lds+

d-

dtack+ d+

dtack-

d- ldtack- ldtack+

dsr- d+

dsr+

ldsd

d- ldtack- ldtack+

dsr- d+

dsr+ lds+Merge similar components

11

Resolving CSC conflicts


dsr- d+

dsr+ lds+

lds-

d-

ldtack-

ldtack+ dsr-d+dsr+ lds+ lds+

dsr+e1 e2 e3 e4 e5 e7

e9

e11

e10

e8

12

Resolving CSC conflicts (cont’d)


dsr- d+

dsr+ lds+csc+

csc-

13

Resulting Circuit

Device

d

Data TransceiverBus

dsr

dtacklds

ldtack

csc

14

Implementation

DESIJ

PUNF

MPSAT

Large STGs(specification)

Medium STGs unfolding-based (exact) tests

Small STGs(components)

synthesis

structural (approximate) tests

decomposition

decomposition

15

Safeness-preserving contractions

• Unfolding is more efficient for safe nets

• Decomposition can create unsafe nets

• Contractions have to preserve safeness

t t t

Example Structural condition

16

Auto-conflicts

• Auto-conflicts appear if too many signals wereremoved

• Backtracking reinserts signals which remove the auto-conflict

• Unnecessary backtracking increases thefinal components

a+ a+

17

Implicit places

• Implicit places are places the absence of tokens in which can never be the sole reason for some transition to be disabled

• Such places can be deleted without changing the behaviour of the STG

• Removing such places is essential for decomposition, because they can cause false alarms for other tests prevent contractions

• Structural test looks for a subset of implicit places (redundant places, shortcut places)

18

Experimental results

• Large trees composed of alternating levels of sequencers and parallelisers were considered

• Intractable for stand-alone MPSAT and Petrify

19


0

500

1000

1500

2000

2500

3000

3500

4000

4500

0 1000 2000 3000 4000 5000Signals

Tim

e[s]

20


• Outperforms stand-alone

MPSAT and Petrify on large STGs• Some intractable for

stand-alone MPSAT and

Petrify benchmarks were

easily synthesized• Huge STGs can be synthesized, e.g.

SeqParTree-10 with 12598 places, 8188 transitions, and 1025 inputs and 3069 outputs was synthesized in less then 70 minutes

21

Thank you!Any questions?

Combining Decomposition and Unfolding for STG Synthesis (application paper) Victor Khomenko 1 and Mark Schaefer 2 1 School of Computing Science, Newcastle.

Documents