B Results on Silicon XOR Arbiter PUFs We also investigated the case of XOR

B results on silicon xor arbiter pufs we also

This preview shows page 13 - 15 out of 16 pages.

B. Results on Silicon XOR Arbiter PUFs We also investigated the case of XOR Arbiter PUFs for FPGA and ASIC data. Our results are summarized in Table X. Again, they are relatively close to our earlier fi ndings of Section IV-A. However, the small deviations from the linear additive delay model now certainly have a stronger effect, since we consider the XOR of several single Arbiter PUFs. We were not able to TABLE X LR ON XOR A RB PUF S OF B ITLENGTH 64 FOR FPGA AND ASIC D ATA (C OLLECTED U NDER S TABLE T EMPERATURE AND M AJORITY V OTING ). T RAINING T IMES A RE A VERAGED O VER D IFFERENT PUF-I NSTANCES Fig. 11. Performance of LR on XOR Arbiter PUFs for FPGA and ASIC data for small prediction errors. learn 6-XOR Arb PUFs anymore with the collected amount of data. Extrapolating from our previous experience, we believe that about 700,000 CRPs would be necessary to this end. 1) Scalability: We also conducted detailed scalability ex- periments, following the methodology of Section IV-B. The required number of CRPs vs. the achieved prediction error is shown in Fig. 11. It shows that for XOR Arb PUFs, the saturation effect is similar to single Arbiter PUFs. The only difference is that it already starts at slightly lower prediction rates, and slowly increases with the number of XORs. Still, the saturation is so mild that also prediction errors below 1% can be achieved, provided that a suf fi cient amount of CRPs is used. Over 1%, the basic relationship (20) appears to hold well, as discussed already in Section IV-B. In terms of computation times, our fi ndings are summarized in Fig. 12. It corresponds to Fig. 4 in Section IV-B, which used simulated CRPs. Again, our results at least qualitatively con fi rm the scaling behavior we earlier observed on simulated data. Also for FPGA and ASIC data, the expected number of restarts to obtain a valid decision boundary on the training set (that is, a parameter set that separates the training set), is given approx- imately by (21) Furthermore, each trial again has the approximate complexity (22)
Image of page 13
RÜHRMAIR et al. : PUF MODELING ATTACKS ON SIMULATED AND SILICON DATA 1889 TABLE XI S OME OF O UR M AIN R ESULTS FOR S IMULATED , N OISE -F REE CRP S AND FOR S ILICON CRP S F ROM FPGA S AND ASIC S . T HE P REDICTION R ATES AND T RAINING T IMES A RE A VERAGED O VER S EVERAL I NSTANCES . A LL P RESENTED T RAINING T IMES A RE C ALCULATED AS IF THE ML E XPERIMENT W AS R UN ON O NLY O NE S INGLE C ORE OF O NE S INGLE P ROCESSOR . U SING C ORES W ILL A PPROXIMATELY R EDUCE T HEM BY Fig. 12. Average rate of success of the LR algorithm on XOR Arbiter PUFs for FPGA and ASIC data, plotted in dependence of the ratio [see (11)] to . X. S UMMARY A. Summary We investigated the resilience of several electrical Strong PUF designs against modeling attacks. To that end, we applied various machine learning techniques to challenge-response sets from two sources: (i) Pseudorandom numeric simulations which used an additive delay model, with and without arti fi cially in- jected errors; and (ii) Silicon CRP data from FPGAs and ASICs.
Image of page 14
Image of page 15

You've reached the end of your free preview.

Want to read all 16 pages?

  • Summer '15

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern

Stuck? We have tutors online 24/7 who can help you get unstuck.
A+ icon
Ask Expert Tutors You can ask You can ask ( soon) You can ask (will expire )
Answers in as fast as 15 minutes
A+ icon
Ask Expert Tutors