edfas.org ELECTRONIC DEV ICE FA I LURE ANALYSIS | VOLUME 24 NO . 4 20 a set of ten SUVs. We set the temperature-controlled chamber to 88oC ± 11oC. Figure 9a shows the degradation plot of the number of stable bits as the IC was aged, and Table 3 shows the stable bit values as the IC was artificially aged. As the IC aged, all SUVs for the five FPGAswere consistently degrading. Figure 9b shows the average percentage of the stable bits as the IC aged. The average number of stable bits for five FPGAs at week-0 was 96.84%, or 5016.31 stable bits. After five years of artificial aging, which was 255 weeks, the average number of stable bits dropped to 60.336%, or 3125.405 stable bits. The average standard deviation of these stable values for all five years of artificial aging was 4.279%. It should be noted that there were enough stable bits available in each IC after artificial aging to create fingerprints. FUTURE WORK The authors are exploring different LUT configurations tomake thememometer PUF a stronger PUF. This includes performing different temperaturemeasurements to analyze thePUF start-upbehavior at various temperatures, and adding error correction to the fingerprints formore robustness over time andunder harshenvironmental conditions. The current methodology can create unique unclonable fingerprints for every FPGA (legacy or contemporary). These fingerprints are further used for authentication within the supply chain. The methodology is being extended to not only passively track these FPGAs but to actively control the ICs by preventing them from further entering the supply chain. Methods that can destroy these ICs when proven untrusted, erase IP and prevent access to the IP are also being explored. CONCLUSION The memometer is a practical hardware metering fingerprint methodology for both legacy and contemporary FPGAs. The authors have developed a new PUF based on cross-coupled LUTs that can overcome manufacturing memory power-on preset. The fingerprints are not only unique but also reliablewith average inter-chip and intrachip HDs close to the ideal 50%and 0%. Instead of having one fingerprint per device, this methodology makes provision for hundreds of fingerprints. There are plans to extend the methodology to make it a stronger PUF, thus making it difficult for an adversary to reverse engineer or clone the fingerprints. MEMOMETER: MEMORY PUF-BASED HARDWARE METERING METHODOLOGY (continued from page 17) Fig. 9 (a) Number of stable SUVs as the FPGA ages (b) average percentage of stable bits as the FPGA ages. (a) (b) Table 3 Artificial aging analysis of memometer PUF on five FPGAs for five years Artificial age (# of weeks) Number of stable bits to a total of 5180 FPGA1 FPGA2 FPGA3 FPGA4 FPGA5 0 5032 5038 5040 4990 4984 1 4755 4626 4557 4371 4465 3 4596 4549 4373 4072 4231 7 4457 4439 4207 3863 4005 15 4251 4189 3984 3596 3711 31 4061 3982 3808 3411 3523 63 3757 3747 3624 3182 3303 127 3532 3466 3463 3000 3153 191 3336 3351 3435 2879 3034 255 3265 3269 3377 2777 2939
RkJQdWJsaXNoZXIy MTMyMzg5NA==