A Robust Protocol for Concurrent OnLine Test COLT of NoCbased SystemsonaChip

1
A Robust Protocol for Concurrent On-Line Test
(COLT) of NoC-based Systems-on-a-Chip

• Design Automation Conference (DAC)
• June 7, 2007

Praveen S. Bhojwani and Rabi N. Mahapatra Texas
AM University, College Station
2
Agenda

• Introduction
• Concurrent On-Line Test (COLT) basics
• Multi TI-IP Configuration
• Robust TI-IP Operation
• Experimental Setup
• Results
• Summary
• Future Work

3
Introduction

• Increasing complexity in SoC designs
• Reliability challenges due to lowering lifetimes
• Due to electro-migration, stress migration,
  time-dependant dielectric breakdown and thermal
  cycling
• Post SoC deployment Run-time confidence in
  correct operation of the SoC
• Precursor to recovery invocation!

4
Possible Options

• Option 1 Use a parallel stream of execution for
  comparison
• Option 2 Turn-off executing applications, put
  SoC into test mode
• Manage test costs
• Test power (up to 1.5x normal mode power)
• Test time (increases with complexity)

5
Possible Options

• Option 3 Run tests while still executing
  applications ? concurrent testing
• Impact on application needs to be managed
• Along with other test costs

6
Networks-on-Chip (NoC)

• Emergence of Networks-on-chip (NoC) to address
  scalability issues with busses for complex
  multi-core designs
• Power efficient
• Reduced contention
• Benefits of interconnection networks!
• Possibility of reuse as a Test Access Mechanism
  (TAM)
• Prevents the need for a design specific TAM

7
On-Line Testing in Research

• Manufacturers insert Infrastructure-IPs (I-IPs)
  into SoC to improve yield and to provide test
  support within designs Zorian,DT02
• Reuse I-IPs to perform the on-line SoC test
• Detect idle periods of execution and test all SoC
  components using I-IPs Marzone,IOLTS05
• non-concurrent on-line testing ? turn off
  applications
• But detecting the idle periods is achallenge
  itself!

8
COLT with a TI-IP

• Concurrent On-Line Testing (COLT) using a Test
  Infrastructure-IP (TI-IP)
• Test in the presence of executing applications
• No need to turn off applications
• Deployed a TI-IP in a NoC-based design to manage
  COLT
• Identified challenges to COLT

P. Bhojwani and R. Mahapatra, An
Infrastructure-IP for on-line testing of
network-on-chip based SoCs, Proc ISQED 2007.
9
Conceptual NoC-enable SoC TI-IP
10
Simplified Example
TR
SSResp
SSResp
SSResp
IP
C
C
C
C
System Snapshot Determination
Test Acceptance Stage
Test Delivery Application Stage
Test Request Stage
SSResp
SSResp
SSResp
SSResp
C
C
C
C
SSReq
SSReq
SSReq
SSReq
SSReq
SSReq
SSReq
SSReq
SSReq
SSReq
SSReq
SSReq
SSReq
SSReq
SSResp
SSResp
SSResp
TV
TV
TV
TV
TI-IP
TI-IP
C
C
C
C
SSResp
SSResp
SSResp
SSResp
C
C
C
C
C Core Network Interface
11
Motivation

• Explore multi TI-IP operation
• Test cases exposed some possible hazards
• Need for a robust operation specification
• Formalize operation
• Specify Protocol
• Identify Hazards
• Provide mitigation support

12
Multi TI-IP Configuration

• For scalability, deploy multiple TI-IPsinto NoC
• Test Vector Delivery cost management

13
Multi TI-IP Operation

• Reduce energy consumption for TV delivery
• Test of TI-IPs
• Challenge
• Co-operative operation to manage COLT
• Manage costs
• Token ring of TI-IP
• TI-IP with the token is the active TI-IPin the
  system

14
Multi TI-IP Operation
Only TI-IP with Test Management Token can accept/
schedule/ execute test requests

• TI-IP

15
Protocol Specification

• Identify communicating partners
• TI-IP
• CNI
• IP core under test
• Identify messages between these partners
• Commands for TI-IP operation
• Test Request, Test Response, System Snapshot
  Determination, System Snapshot Response, Test
  Vector Data, Test Throttle,CNI Alarm,

16
Test Request Protocol Step
CNI attached to IP Core to be tested
17
System Snapshot Collection Step
System Snapshot Requests sent to all CNIs in
the NoC
18
Test Vector Delivery
CNIs attached to cores-under-test
Single TI-IP setup
19
Test Throttle Step
Test Throttle Request only sent to necessary CNIs
of IP cores under test
20
Test Management Token Step
3 TI-IP setup
21
Protocol Hazards

• Identify design aspects that may affectCOLT
  operation
• Done for each protocol step
• Identified 5 types of hazards
• Starvation due to Test/Application
• Test Input Queue Buffer Overflow
• TI-IP Failure
• Test Wrapper Buffer Overflow
• CNI Failure

22
Protocol Hazards

• Starvation
• Application and Test traffic, prevents theother
  from communication
• Test Input Queue Buffer Overflow
• Excessive test traffic can lead to buffer
  overflow at the TI-IP input
• TI-IP Failure
• CNI Failure

23
Protocol Hazards

• Test Wrapper Buffer Overflow
• Can lead to test data loss

24
Hazard Summary
25
Hazard Mitigation

• Mitigate the identified hazard for each protocol
  step
• Mitigation techniques used
• Timeouts
• Priority inversions
• Communication interleaving
• Automatic Retransmission
• Some failures cannot be recovered from!
• Details in the paper

26
Experimental Setup

• Simulation platform
• NoCSim NoC interconnection network simulator
• 4x4 2D folded torus
• Application Benchmarks
• Embedded System Synthesis Benchmark Suite (E3S)
• Task graphs from 5 application domains
• Test Benchmarks
• ITC02 SoC Test Benchmark
• g1063.soc and d695.soc test cases used

27
Experimental Setup

• Only considering SCAN for now due to test data
  availability
• Only testing IP cores (for now)
• Assign 1 TI-IP to an SoC and let it occupy a
  whole NoC tile
• For a 22mm x 22mm chip laid out as a 4x4 2D
  torus, each tile could be 5mm x 5mm Towles,
  DAC01
• 5.2 area overhead

28
Test Configurations
Each network tile configured for an IBM 405GP
(area constraints)
Task graph assignment done by hand.
29
Test Configurations

• Multi TI-IP test configurations tested and
  operation verified
• Protocol hazard scenarios developed

30
Results
Normal on line test-mode vs Starvation Hazard
test mode energy profile
6
.
00
E
-
09
5
.
50
E
-
09
5
.
00
E
-
09
)
J
(

4
.
50
E
-
09
y
g
r
4
.
00
E
-
09
e
n
E
3
.
50
E
-
09
3
.
00
E
-
09
2
.
50
E
-
09
0
9
8
7
6
5
4
3
2
1
0
9
8
3
5
7
9
1
3
5
1
7
9
0
2
2
3
4
5
7
8
9
1
0
1
3
4
1
1
1
1
Time
Normal on
-
line test mode
Mitigating starvation impact on test
31
Results
32
Summary

• Lowering design lifetimes
• Concurrent On-Line Testing techniques needed
• Use of TI-IPs to provide COLT support
• Multi TI-IP configuration to manage test vector
  delivery cost
• Robust TI-IP operation essential
• Protocol Specification
• Hazard Identification
• Hazard Mitigation

33
Summary

• Verified TI-IP and multi TI-IP operation using
  combination of academic benchmarks
• Developed test cases to generateHazard scenarios
• Verified Robust TI-IP operation underthese
  scenarios

34
Future Work

• Explore test triggering options
• Event based triggering
• On-line testing of NoC components
• Routers
• Links
• CNI

35

• Questions?