Baloch 1 MAPLD 2005/1024

Design of a ‘Single Event Effect’ Mitigation Technique for

Reconfigurable Architectures

TH

E

UNI VERSI TY

OF

E D I N B U RG

H

SAJID BALOCH

Prof. Dr. T. Arslan1,2 Dr.Adrian Stoica3Supervisory Team

Baloch 2 MAPLD 2005/1024

ACRONYMES

– SEU (Single Event Effect)– SET (Single Event Transient)– SEB (Single Event Burnout)– SEL (Single Event Latch-up)– Cfg (Configuration)– EDAC (Error Detection and Correction)– SoC (System on Chip)– FPGA (Field Programmable Gate Array)– DEU (Double Event Upset)– TEU (Triple Event Upset )– MEU (Multiple Event Upsets)

Baloch 3 MAPLD 2005/1024

RECONFIGURABLE ARCHITECTURES

ME

MO

RY

ME

MO

RY

ReconfigurableArray

IP

DSP

IP

Re-Configurable SoC Architecture

a) FPGAs - SRAM - Anti Fuse - EPROM

b) Reconfigurable SoC - General purpose - Domain Specific

Baloch 4 MAPLD 2005/1024

RADIATION EFFECTS RE-CONFIGURABLE ARCHITECTURES

– PERMANANT FAULTS (due to SEL, SEB etc)

– TEMPORARY FAULTS (due to SEU etc)

Baloch 5 MAPLD 2005/1024

SEU MITIGATION TECHNIQUES

a) HARDWARE REDUNDANCY - Dual Modular Redundancy (DMR) - Triple Modular Redundancy (TMR) - EDAC Codes - Process Technology

b) TIME REDUNDANCY

c) COMBINATION (Hardware & Time)

Baloch 6 MAPLD 2005/1024

• TRANSIENT FAULTS (Data Memory etc)

• PERMANANT FAULTS (Cfg. Memory)

Radiation Hardening SEU EFFECTS

Baloch 7 MAPLD 2005/1024

SEU EFECTS Synchronous Circuits

D-F

lip F

lop

D-F

lip F

lop

CombinationalLogic Circuits

Data_in Data_out

Clock

Baloch 8 MAPLD 2005/1024

SEU EFECTS Configuration Memory

M

M

D-F

lip F

lop

M

M M M MLUT

SEU (BITFLIP)

Baloch 9 MAPLD 2005/1024

CLUSTER

CLUSTER

CLUSTER

CLUSTER

CLUSTER

CLUSTER

SEU EFECTS ROUTING OF A SIGNAL

Baloch 10 MAPLD 2005/1024

Proposed SEU/SET MitigationTechnique

based on:• Temporal Data Sampling• Weighted Voting

Salient Features of The Proposed Technique:Auto Correction Mechanism for• 100% SEU Recovery• 100% Double Fault Recovery• Voter Faults Recovery

Baloch 11 MAPLD 2005/1024

Temporal Sampling

Q

QSET

CLR

S

R

Q

QSET

CLR

S

R

Q

QSET

CLR

S

R

Q

QSET

CLR

S

R

Q

QSET

CLR

S

R

Q

QSET

CLR

S

R

Data

CLk-A

CLk-B

CLk-C

L1 L2

L6L5

L4L3

1

3

5 6

4

2

Primary Section

Secondary Section

Baloch 12 MAPLD 2005/1024

TEMPORAL SAMPLING Clock Scheme

• 3 derivates of Main Clock

• Each Clock is Phase shifted

• 25% duty Cycle

CLk-A

CLk-B

CLk-C

CLOCK

Computation Cycle

2 Clock Cycles

Baloch 13 MAPLD 2005/1024

Minimized TermX4.X3.X0 + X5.X3.X0 + X5.X4.X0 + X4.X3.X1 + X5.X3.X1 + X5.X4.X1 + X4.X3.X2 + X5.X3.X2 + X5.X4.X2 + X5.X4.X3 +

X3.X2.X1.X0 + X4.X2.X1.X0 + X5.X2.X1.X0

Weighted Voter Circuit

Baloch 14 MAPLD 2005/1024

SEU in Secondary SectionNode Before

SEUAfterSEU

Voting Weights‘1’ ‘0’

Total Votes ‘1’ ‘0’

1 1 1 2 -

8 1

3 1 1 2 -

5 1 1 2 -

2 1 0 - 1

4 1 1 1 -

6 1 1 1 -

Case Example

Baloch 15 MAPLD 2005/1024

Multiple Bit Upset

Node BeforeSEU

AfterSEU

Voting Weights‘1’ ‘0’

Total Votes ‘1’ ‘0’

1 1 1 2 -

5 4

3 1 0 - 2

5 1 0 - 2

2 1 0 1 -

4 1 1 1 -

6 1 1 1 -

Case Example

Baloch 16 MAPLD 2005/1024

Hardware Implementation of Proposed Scheme with

Auto-Correction Mechanism

Baloch 17 MAPLD 2005/1024

Single Event Transition FaultData / Clock

Baloch 18 MAPLD 2005/1024

SEU/SET Simulator•SEU’s can be injected at instance•SEU of any duration can be injected•Multiple upsets can be injected

Baloch 19 MAPLD 2005/1024

Performance AnalysisFault Coverage

 Mitigation Scheme %age Fault ToleranceSET SEU DEU TEU

Proposed Scheme 100% 100% 100% 50%F. Lima etal Scheme 63% 100% - -D. Mavis etal Scheme 100% 100% 32% 18%

Baloch 20 MAPLD 2005/1024

Performance AnalysisArea Overhead

0

50

100

150

200

250

300

350

400

450

Proposed Technique Mavis etal Technique

Area

(u s

quar

e m

eter

)

Results are based on:

0.13µm CMOS technology