EVALUATION OF PHYSICAL UNCLONABLE FUNCTIONS

Transcription

1 EVALUATION OF PHYSICAL UNCLONABLE FUNCTIONS ECE 646 PROJECT PRESENTATION DEC 11, 2012 YAMINI RAVISHANKAR

2 PHYSICAL UNCLONABLE FUNCTIONS A challenge-response mechanism in which the mapping between a challenge and the corresponding response is dependent on the complex and variable nature of the physical material Unique challenge-response pairs for each chip

3 BIOMETRICS OF INTEGRATED CIRCUITS Identify individual PUF response Identify Integrated Circuit

4 WHY UNCLONABLE? Physical Unclonable Mathematical

5 TYPES OF PUFS Silicon Arbiter PUF Ring Oscillator based PUF Butterfly PUF SRAM based PUF Non-Silicon Coating PUF Magnetic PUF RF PUF

6 RING OSCILLATOR PUF r = 1 if Freq(ROi) > Freq(ROi+1) else r = 0

7 NOTATIONS USED N = total number of chips analyzed n = index of a chip (range 1 to N) M = total number of ring oscillators K = total number of identifiers(ids) generated per chip k = index of an ID in a chip (range 1 to K) T = total number of samples measured per ID t = index of a sample (range 1 to T) R = L-bit response from the PUF r = l th bit of the response R

8 DATA SETS COMPARED GMU Virginia Tech N (no of chips) M (no of ROs) K (no of Identifiers) L (length of response) T (no of samples) in configurable mode 512 in normal mode

9 CONFIGURABLE RING OSCILLATOR Each RO consists of 8 frequencies such as a look-up table GMU 2 datasets Horizontal scanning Vertical scanning

10 NEED FOR METRICS To evaluate the performance of different types of PUFs To be able to compare one PUF with the other To standardize the security requirements expected from the PUF

11 METRICS ANALYZED Uniqueness Bit-Aliasing Uniformity Randomness Reliability Correctness Steadiness Domains specified by Maiti et al. for the metrics

12 UNIQUENESS (DEVICE) Average inter-chip Hamming distance (HD) computed across a group of chips According to Maiti et al., Ideal value = 50% According to Hori et al., Ideal value = 100% Estimate of the inter-chip variation in terms of PUF responses

13 BIT ALIASING (DEVICE) Hamming weight of the l-th bit of the identifier across N devices Bit-aliasing is given by Ideal value = 50% Estimating the bias of a particular response bit across several chips Gives information about any systematic and spatial effect across several devices

14 UNIFORMITY (SPACE) Proportion of 0 s and 1 s in the response bits of a PUF Uniformity is given by Ideal Value = 50% Uniformity is the percentage Hamming Weight of the l- bit response Calculated for each chip individually from one response

15 RANDOMNESS (SPACE) Randomness indicates the balance of 0 and 1in the response bits of the PUF (Hori 2010) Randomness is given by Ideal value = 100% Similar to the uniformity parameter defined by Maiti et al. Takes into consideration T samples of the response bits

16 RELIABILITY (TIME) How reliable are the response bits at varying operating conditions Reliability is given by Ideal value = 100% Calculated based on the average value of the intrachip Hamming Distance Calculated over T samples for each chip

17 CORRECTNESS (TIME) Sum of the Hamming Distances (SHD) normalized by T, K and L Correctness is given by Ideal value = 100% Similar to the reliability parameter defined by Maiti et al. Can be correlated with the reliability parameter as Correctness = (2xReliability) - 1

18 STEADINESS (TIME) Degree of bias of a response bit towards 0 or 1 over T samples Steadiness is given by Ideal value = 100% Bias of individual response bits on an average Calculated for each individual bit in a response over T samples Does not take time factor into consideration

19 PARAMETERS NOT ANALYZED Diffuseness (space) Degree of difference among the IDs generated from the different challenge sets in the same device More than one identifier required to calculate the value of diffuseness Probability of Misidentification (device) Rate of error in the identification of a chip by a PUF due to noise in the response bits

20 CALCULATED METRICS Parameter Uniqueness Maiti Uniqueness Hori Ideal value (in %) Maiti thesis VT data (in %) VT Calculated values (in %) GMU - Horizontal GMU - Vertical Bit-aliasing Uniformity Randomness Reliability Correctness Steadiness

21 COMPARISON PLOTS Ideal Value Maiti thesis VT GMU - Horizontal GMU - Vertical 0

22 RESULTS AND ANALYSIS GMU PUF exhibits better reliability, steadiness and correctness in comparison with the VT data Uniformity and bit-aliasing of GMU PUF responses is closer to the ideal value than the VT responses Randomness is almost the same in both cases The uniqueness in the GMU data is very low. This can be attributed to the continuous strings of 0 s and 1 s in the identifiers The uniqueness of the GMU data can be improved by performing post-processing operations like Identity- Mapping and Quantization such that continuous strings of 1 s and 0 s are avoided

23 TIME SCHEDULE Oct Literature survey on the metrics available for evaluating PUFs Oct Familiarized with the scripting language Python to analyze the data Nov Worked on the GMU and VT data available and calculated the various metrics Nov Comparative analysis of the data Dec 11 Final Presentation

24 REFERENCES A. Maiti and P. Schaumont, Improved Ring Oscillator PUF: An FPGA-friendly Secure Primitive, Journal of Cryptology, DOI: /s Hori et.al., Quantitative and Statistical Performance Evaluation of Arbiter Physical Unclonable Functions on FPGAs, ReConfig2010, Pg

25 Thank you