Soft Response Generation and Thresholding Strategies for Linear and - - PowerPoint PPT Presentation

International Symposium on Low Power Electronics and Design

Soft Response Generation and Thresholding Strategies for Linear and Feed-Forward MUX PUFs

Chen Zhou, Saroj Satapathy, Yingjie Lao, Keshab K. Parhi and Chris H. Kim

Department of ECE University of Minnesota

2

Outline

• Physical Unclonable Function (PUF)
• 32nm PUF Chip Measurements
• Soft Response Thresholding Strategies
• Linear PUF vs. Feed-forward PUF
• Conclusion

• Unique and random:

Based on inherent process variation

• Secure: Large # of

challenge-response pairs (CRPs)

3

PUF Fingerprint of chip

Inputs Outputs

Numerous input choices

Challenges Chip #1

Unique and random responses

Challenge #1 Challenge #2 Challenge #n Response #1 Response #2 Response #n

Chip #2 Responses #1 Responses #2

Physical Unclonable Function (PUF)

• Server-user based authentication
• Challenge-response pairs tested and stored before

usage

4

Typical Authentication Process

Database

Chip PUF ID

Server User Store CRPs

• f all PUFs

Public Private

• Public chip ID is first sent to the server

5

Typical Authentication Process

• Server retrieves CRP subset table for the given chip ID

6

Typical Authentication Process

Retrieve CRP subset Database

Chip PUF ID Challenge 209A (Hex) 41B1 (Hex) 9283 (Hex) Response 1 1

Server User

• Challenges are sent to the user

7

Typical Authentication Process

Retrieve CRP subset Database

Chip PUF ID Challenge 209A (Hex) 41B1 (Hex) 9283 (Hex) Response 1 1

Server User Challenges

• User generates responses using PUF circuit

8

Typical Authentication Process

Database

Chip PUF ID Challenge 209A (Hex) 41B1 (Hex) 9283 (Hex) Response 1 1

Server User

Response 1 1

• User responses are sent to server for comparison

9

Typical Authentication Process

Database

Chip PUF ID Challenge 209A (Hex) 41B1 (Hex) 9283 (Hex) Response 1 1

Server User

Compare Response 1 1

Responses

• Approved if responses match; denied if mismatch
• Final step: decision sent to user

10

Typical Authentication Process

Denied

Database

Chip PUF ID Challenge 209A (Hex) 41B1 (Hex) 9283 (Hex) Response 1 1

Server User

Compare Response 1 1

Decision match mismatch

Approved

• Hamming distance can be used as matching criteria
• Intra-chip HD: Same chip, noise effects, close to 0%
• Inter-chip HD: Different chip, process variation effects,

close to 50%

11

Hamming Distance (HD) Calculation

12

Outline

• Physical Unclonable Function (PUF)
• 32nm PUF Chip Measurements
• Soft Response Thresholding Strategies
• Linear PUF vs. Feed-forward PUF
• Conclusion

13

Motivation of This Work

Hamming distance

PDF

0.5

Intra-chip Inter-chip

1.0 0.0

Ideal case (all CRPs)

No overlap

Hamming distance

PDF

0.5

Intra-chip Inter-chip

1.0 0.0

Actual case (only stable CRPs)

Hamming distance

0.5

Intra-chip Inter-chip

1.0 0.0

Actual case (all CRPs)

PDF

No overlap Overlap

• Stable CRPs have less

intra-chip variation

• Measure soft response

(=probability of response being ‘1’ or ‘0’) to find stable CRPs

14

Contributions of This Work

• Implemented soft response collection circuits in a

32nm test chip

• Generated MUX PUF soft response distribution based
• n 3.3 Gb test data
• Proposed soft response thresholding strategies to

select stable challenge-response pairs

• Implemented and characterized feed-forward MUX

PUF

15

Proposed Soft Response Measurement Circuit

• Soft response = response probability information
• >GHz sampling circuits facilitate efficient soft response

measurements

• Parallel or crossed signal paths

configured by challenge bits

• Delay difference determined

by inherent process variation

16

Linear MUX PUF Delay Stages

1 PUF stage (c='1') PUF stage (c='0')

• Arbiter generates

response bit based on delay difference

17

Arbiter Circuit

c1

On- chip VCO

c2 c32

32 MUX stages Arbiter On-chip counter SR Latch S Q R On-chip counter M Response N

Δ

18

32nm PUF Test Chip

19

Soft Response Measurements

• Soft response is a function of the actual delay

difference

• Above distribution generated using 3.3 Gb of PUF

response data

20

Outline

• Physical Unclonable Function (PUF)
• 32nm PUF Chip Measurements
• Soft Response Thresholding Strategies
• Linear PUF vs. Feed-forward PUF
• Conclusion

21

• Symmetric thresholds used to define stable and

unstable CRPs

• Unstable CRPs not used for authentication

Soft Response Thresholding Strategy

22

• Left: HD distributions overlap when threshold=0.5
• Right: No overlap when threshold=0 and 1 (i.e. only

stable responses are used)

Impact of Soft Response (SR) Threshold

23

Fixed Threshold Scheme

• No stable ‘1’ to stable ’0’ flips when threshold > 0.81
• Stable ‘1’ to ‘unstable’ flips always exist, necessitating

more tests to find stable CRPs

24

Relaxed Threshold Scheme

• Stringent threshold during enrollment phase and

relaxed threshold during authentication

• Results in fewer ‘1’
• ’unstable’ and ‘0’
• ’unstable’ flips

25

Outline

• Physical Unclonable Function (PUF)
• 32nm PUF Chip Measurements
• Soft Response Thresholding Strategies
• Linear PUF vs. Feed-forward PUF
• Conclusion

26

Linear MUX PUF Vulnerability

= 2 − 1)(2 − 1) ⋯ (2 − 1 (2 − 1) ⋯ (2 − 1) ⋮ (2 − 1) 1

• = 1

2

• −
• +

+ −

• ⋮
• +

+ −

• +

+

∆ = · = ( ∆ + 1 ) 2 ⁄

c1 c2 c32

32 MUX stages Arbiter SR Latch S Q R Response

Δ

• Linear PUFs are susceptible to modelling attack
• That is, attacker can predict correct response with very

high probability using past CRP data

27

Feed-forward MUX PUF for Improved Security

• Use intermediate response for some challenge bits
• Non-linear relationship between delay and response
• harder for attacker to predict correct response
• No experimental data reported on feed-forward PUF

Feed-forward MUX PUF ref.: J. W. Lee, et al., VLSI Circuits Symposium, 2004

28

32nm Test Chip Data: Linear vs. Feed- forward MUX PUF

• % of stable CRPs decreases from 94.16% to 91.02%

due to instability of internal challenge bit

29

Conclusion

• Soft response measurement circuit demonstrated in a

32nm test chip

– On-chip VCO and counters enable fast measurement

• Different thresholding strategies evaluated

– Enables robust authentication across wider voltage and temperature range

• Feed-forward MUX measured for the first time

– % of stable CRPs decreases slightly due to instability of internal challenge bit

Acknowledgements

• National Science Foundation and Semiconductor

Research Corporation for funding