1 Valentin Muresan©

Valentin Muresan,Xiaojun Wang

Dublin City UniversityValentina Muresan,

Mircea Vladutiu“Politehnica” University of

Timisoara

2 Valentin Muresan©

Introduction

� Block-test scheduling problem having assigned power dissipation limits

� Adapted classical high-level synthesis scheduling approaches

� Extended tree growing heuristic

� Constant additive power dissipation model

3 Valentin Muresan©

Why Do This?

� To solve the problem of test scheduling having power dissipation limits

� To propose a practical approach for the power-constrained test scheduling (PTS)

� To achieve a balanced test-power dissipation

� To improve the test concurrency with power dissipation limits

� To generate a near-optimal PTS solution under loose power dissipation constraints

4 Valentin Muresan©

Outline

� Introduction� Proposed approach� Extended compatibility tree example� Power-test scheduling chart example� Power-test scheduling example� Adapted HLS approaches� Power-test scheduling charts (comparison)� Power-test characteristics vs. power limits� Power-test distribution graph� Experimental results� Advantages vs. disadvantages� Conclusions

5 Valentin Muresan©

Proposed Approach

�Unequal-length block-test scheduling�Partitioned testing with run to completion�Constant additive model

� Assumes constant power dissipation value for each block-test

•Average power dissipation•Maximum power dissipation•RMS power dissipation

� Power dissipation of a test session is the sum of the block-tests’ power dissipations

6 Valentin Muresan©

Example of Block Under Test

7 Valentin Muresan©

Extended Tree Growing Approach (A)

�Merging step example

8 Valentin Muresan©

Extended Tree Growing Approach (B)�Test scheduling chart and extended

compatibility tree example

9 Valentin Muresan©

Power-test Scheduling Chart Example

10 Valentin Muresan©

Characteristics of the Power-test Scheduling Chart

11 Valentin Muresan©

Power-test Scheduling Example

•Stepwise PTS example•Stepwise tree growing•PTS-LEAFDS approach

•t1and t4 are sequential•t1and t2 are in parallel

Valentin Muresan©

Adapted HLS Approaches

�Left-edge algorithm based approach (PTS-LEA)

�List scheduling based approach (PTS-LS)�Force-directed scheduling based approach

(PTS-FDS)�Mean-square error based approach (PTS-

MSE)�A combination (sequence) of list and force-

directed scheduling approaches

12

Valentin Muresan©

Power-test Scheduling Chart (A)�List scheduling based PTS-LEA approaches

13

Valentin Muresan©

Power-test Scheduling Chart (B)�PTS-LEAFDS and PTS-FDS approaches

14

Valentin Muresan©

Power-test Scheduling Chart (C)�PTS-LEA approach with power constraints

15

Valentin Muresan©

Power-test Scheduling Chart (D)� PTS-LEA approach with power constraints

16

Valentin Muresan©

Power-test Characteristics for PTS-LEA Approach Over the Power Dissipation Constraints Range (A)

17

Valentin Muresan©

Power-test Characteristics for PTS-FDS Approach Over the Power Dissipation Constraints Range (B)

18

Valentin Muresan©

Power-test Characteristics for PTS-LEAFDS Approach Over the Power Dissipation Constraints Range (B)

19

Valentin Muresan©

Experimental Results

PTS-LEA

PTS-LEA

PTS-FDS

PTS-MSE

PTS-LEAFDS

PTS-LEAMSE20

Valentin Muresan©

Experimental Results (Conclusions)

�PTS-LEA suitable for giving solutions faster with good test application times

�PTS-FDS gives solutions with much more balanced power dissipation distribution

�PTS-LEAFDS combines (for loose power constraints) the above good results

�PTS-LEAFDS is computationally slightly more expensive than PTS-FDS

�For tighter power constraints the test application time of the PTS-LEA solutions can not be accomodated anymore by PTS-LEAFDS

21

Valentin Muresan©

Conclusions

�Proposed a combined approach for the non-polynomial problem of PTS

�This approach achieves quickly PTS results for designs ranging from RTL to system-level

�The PTS problem has been modeled as an extended tree growing approach

�Classical scheduling approaches have been combined (sequenced)

�Advantages/disadvantages of the PTS approaches have been compared

22