Upload
others
View
6
Download
0
Embed Size (px)
Citation preview
ARO-PUF: An Aging-ResistantRing Oscillator PUF Design
Md. Tauhidur Rahman1, Domenic Forte1, Jim Fahrny2, and Mohammad Tehranipoor1
1ECE Dept., University of Connecticut 2Comcast
{tauhid, forte, tehrani}@engr.uconn.edu jim [email protected]
Abstract—Physically Unclonable Functions (PUFs) haveemerged as a security block with the potential to generate chip-specific identifiers and cryptographic keys. However it has beenshown that the stability of these identifiers and keys is heav-ily impacted by aging and environmental variations. Previoustechniques have mostly focused on improving PUF robustnessagainst supply noise and temperature but aging has been largelyneglected. In this paper, we propose a new aging resistant designfor the popular ring-oscillator (RO)-PUF. Simulation resultsdemonstrate that our aging resistant RO-PUF (called ARO-PUF) can produce unique, random, and more reliable keys. Only7.7% bits get flipped on average over 10 years operation periodfor an ARO-PUF due to aging where the value is 32% for aconventional RO-PUF. The ARO-PUF shows an average inter-chip HD of 49.67% (close to ideal value 50%) and better thanthe conventional RO-PUF (∼ 45%). With lower error, ARO-PUFoffers ∼ 24X area reduction for a 128-bit key because of reducedECC complexity and smaller PUF footprint.
Keywords—PUF, reliable RO-based PUF, reliable PUF, robustPUF, aging resistant PUF, PUF reliability.
I. INTRODUCTION
In the effort to design secure systems, Physical Unclon-able Functions (PUFs) have emerged as a potential securityblock [1-4] as they can generate volatile secret keys for asystem. PUF is an integrated circuit that generates secretkeys by exploiting the inherent physical variations of devices.PUFs provide a high level of protection in security block.It offers strong volatile key storage to make the systemtamper-resistant. Functionally, a PUF maps a challenge, inputto the PUF, to responses, PUF output. A PUF exploits thephysical characteristics of silicon and provides an alterna-tive to conventional digital signature stored in non-volatilememory. It uses the physical properties of each device togenerate a chip-specific fingerprint or key. Identical PUFswith the same manufacturing process provide a different setof data because of small variations between each device.PUF can be used extensively in security applications such asIC identification/authentication, hardware metering, certifiedexecution, and key generation for encryption [4-9]. SeveralPUFs have been proposed over the past decade, such as ArbiterPUF [2], ring-oscillator-based PUF (RO-PUF) [4], SRAM-based PUF [13], Butterfly PUF [14], etc. Among them, theRO-PUF is considered more reliable under a wide range oftemperatures and it has been more widely accepted because itis less complex and easier to fabricate [4, 17, 22].
In order to be useful in practical security applications, thefingerprint or key generated by the PUF should be reliable orstable (i.e., not change over time). For example, in cryptogra-phy, decrypting the original message will be impossible withwrong private/public keys. Authentication of devices, securecommunication, etc. also suffer from complications due toPUF instability. It is relatively well known that aging andenvironmental variations lead to performance degradation and
lower reliability in ICs [10-12, 23-24, 32]. Similarly PUFs,hence, the keys they generate will therefore also be renderedless reliable by such variations. Run-time aging effects suchas negative bias temperature instability (NBTI), hot carrierinjection (HCI), oxide breakdown, and electromigration (EM)are the most critical causes of continuous performance andreliability degradation [10]. Threshold voltage (VT ), channellength, timing, etc. are the critical parameters that decide theperformance of a chip. NBTI can degrade VT of a transis-tor by 10-15% in the first year depending on technologyand workload [23], which reduces drive current and hencecircuit performance. During negative bias (pMOS is ON),pMOS experiences a constant degradation. However, partiallyrecovery occurs when the stress is removed (pMOS is OFF).HCI is a function of switching activity and can acceleratedegradation even further by shifting the threshold voltage ofnMOS transistors over time [12]. While HCI mostly impactsthe VT of the nMOS, it may also degrade the drain currentof both nMOS and pMOS [32]. The delay degradation due toNBTI and HCI follows the same trend as the VT degradation.The severity of the degradation due to NBTI and HCI isdependent on workload, logic function, temperature, supplyvoltage, threshold voltage, technology and geometry [10, 21,23-24, 32]. In general, low VT transistors experience moredegradation over time than high VT transistors. Similarly,higher-supply voltage, VDD, and higher-operating temperatureexpedite the delay degradation over time. Fundamentally,aging will permanently shift many critical parameters thePUF uses to generate its key and therefore is a significantconcern to PUF reliability. Another issue for PUF reliabilityis temperature variation. Operating temperature affects thedelay of a device by changing its mobility and thresholdvoltage [12]. As temperature increases, the threshold voltagedecreases which leads to an increase in the drain saturationcurrent. At the same time, it decreases the mobility of theMOSFET, which causes a decrease in the drain saturationcurrent. However, the mobility degradation dominates, andconsequently, the delay of the device decreases. Similar toaging, these temperature-induced changes to delay will alsocreate noise in the PUF’s key. However, unlike aging, theeffects can be reversed by returning the device to its nominaloperating temperature. Like temperature, power supply noisealso affects the delay, because of deviation from nominal VDD,of logic gates and hence the reliability of an IC. The powersupply delivery network and parasitic impedance of packagesresult in VDD fluctuation and impact the IC performance. Likeaging and temperature variation, VDD fluctuation is also liableto impact the reliability of a PUF. Unlike aging the supplynoise’s impact is temporary, i.e., when VDD is back to itsnominal value, the PUF operates more reliably. Researchershave proposed several techniques to improve the reliabilityof a PUF [16-19, 21, 23, 24, 28-29]. For example, errorcorrecting codes (ECC) have been used to fix up to a certainnumber of erroneous bits in the PUF key. For example, ithas been reported in [15] that the index-based syndromefollowed by a 3x repetition code and BCH [30] code can lowerthe RO-PUF error rate below 1 ppm under severe operating
Table IV: % of ′1′ in response to all possible challenges atdifferent temperatures
Temperature (◦C)% ’1’ in response
% ImprovementARO-PUF RO-PUF
−25 60 66 9.090 59 64 7.81
25 55 60 8.3350 58 62 6.4575 58 62 6.45100 58 61 4.91125 58 60 3.33
and 45% ′0′ on average, which is close to the ideal value (50%′1′ and 50% ′0′) and better than the conventional RO-PUF.
Table V: Area overhead comparison for 128-bit key
Blocks ARO-PUF RO-PUF ReductionRequired ROs 510 16382 32xArea for ROs 510*6.5*AnMOS 16382*3.5*AnMOS 17.3xECC Encoder ∼ 3Kgates ∼ 12 Kgates ∼ 4xECC Decoder ∼ 20 Kgates ∼ 75 Kgates ∼ 3.75xAnMOS: Area of nMOS with Wn=0.12u3.5 ≡ (Wn = 0.12u)+(W p = 2.5∗Wn)6.5 ≡ (Wn = 0.12u)+(W p = 2.5∗Wn)+(WMb =Wn)+(WMc = 2∗Wn)
Area Overhead: Table V shows the estimated area overheadcomparison between ARO-PUF and RO-PUF for 128-bit keygeneration based on [30-31]. A system with the ARO-PUFrequires a smaller PUF footprint. For example, to correct∼ 7.9% and ∼ 25% errors for a 128-bit key generation, BCHrequires 255 and 8191 bits respectively [30]. Though AROconsumes more area than a conventional RO, ARO-PUF saves∼ 17x area because of smaller PUF footprint. At the sametime, both BCH encoder and decoder required lower overhead(timing, power and area) since ARO-PUF has less errors thanRO-PUF. Estimation shows that ARO-PUF offers ∼ 7x areareduction for BCH encoder and decoder. ARO-PUF offersmore area efficient implementation for longer key size.
V. CONCLUSION
In this work, we have proposed the first aging-resistantRO-PUF design. The simulation results show that the proposedARO-PUF ages significantly less compared to the conventionalRO-PUF. The ARO-PUF provides a more stable key over longtime and therefore reduces ECC complexity and offers smallerPUF footprint, resulting in a more efficient implementation. Atthe same time, the keys generated by the ARO-PUF are lesssensitive to temperature and also possess strong security prop-erties. In future work, we shall evaluate the proposed ARO-PUF in real ASICs and investigate complementary approachesfor further reducing ECC overhead.
VI. ACKNOWLEDGMENT
This work was supported in part by the National ScienceFoundation under grant CNS-0844995 and a grant from Com-cast.
REFERENCES
[1] R. Pappu, “Physical one-way functions,” Phd thesis, MassachusetsInstutute of Tecnhology, 2001.
[2] B. Gassend et al., “Silicon physical random functions,” in CCS 02: Pro-ceedings of the 9th ACM conference on Computer and communicationssecurity. New York, NY, USA: ACM, pp. 148-160, 2002.
[3] D. Lim et al., “Extracting secret keys from integrated circuits,” IEEETrans. VLSI Syst., vol. 13, no. 10, pp. 1200-1205, 2005.
[4] G. E. Suh and S. Devadas, “Physical Unclonable Functions for DeviceAuthentication and Secret Key Generation,” in Proc. 44th ACM/IEEEDesign Automation Conf. DAC 07, pp. 9-14, June 2007.
[5] F. Koushanfar and M. Potkonjak, “CAD-based Security, Cryptography,and Digital Rights Management,” Design Automation Conference, 44thACM/IEEE, pp. 268-269, June 2007.
[6] J. Guajardo et al., “ Anti-counterfeiting, key distribution, and key storagein an ambient world via physical unclonable functions”, InformationSystems Frontiers, vol. 11, no. 1, pp. 19-41, 2009.
[7] M. Areno, and J. Plusquellic, “Securing Trusted Execution Environmentswith PUF Generated Secret Keys,” in Trust, Security and Privacy inComputing and Communications (TrustCom), pp. 1188-1193, June 2012.
[8] S. Devadas, E. Suh, S. Paral, R. Sowell, T. Ziola, and V. Khandelwal,“Design and implementation of puf-based ”unclonable” rfid ics for anti-counterfeiting and security applications,” in RFID, pp. 58-64, April 2008.
[9] G. E. Suh, C.W. O”Donnell, and S. Devadas, “Aegis: A Single-ChipSecure Processor,” Design & Test of Computers, IEEE , vol.24, no.6,pp. 570-580, Nov.-Dec. 2007.
[10] D. Lorenz, G. Georgakos, and U. Schlichtmann, “Aging analysis ofcircuit timing considering nbti and hci,” in On- Line Testing Symposium,(IOLTS). 15th IEEE International, pp. 3-8, 2009.
[11] F. Catthoor, P.Raghavan, H. Kukner, S. Khan, and S. Hamdioui,“Incorporating parameter variations in BTI impact on nano-scale logicalgates analysis,” IEEE International Symposium on Defect and FaultTolerance in VLSI and Nanotechnology Systems (DFT), pp. 158-163,2012.
[12] I.M. Filanovsky and A. Allam, “Mutual Compensation of Mobilityand Threshold Voltage Temperature Effects with Applications in CMOSCircuits,” IEEE Transactions on Circuits and Systems - 1: FundamentalTheory and Applications, vol. 48, no. 7, pp. 876-884, Jul. 2001.
[13] J. Guajardo, S.S. Kumar, GJ. Schrijen, and P. Tuyls, “FPGA IntrinsicPills and Their Use for IP Protection,” Workshop on CryptographicHardware and Embedded Systems (CHES), Sep. 2007.
[14] S. Kumar, J. Guajardo, R. Maes, GJ. Schrijen and P. Tuyls, “TheButtefly PUF: Protecting IP on every FPGA,” In IEEE InternationalWorkshop on Hardware Oriented Security and Trust, 2008.
[15] M. Yu, and S. Devadas, “Secure and Robust Error Correction forPhysical Unclonable Functions,” Design & Test of Computers, IEEE, vol.27, no.1, pp. 48-65, 2010.
[16] V. Vivekraja, and L. Nazhandali, “Circuit-level techniques for reliablePhysically Uncloneable Functions,” IEEE International Workshop onHardware-Oriented Security and Trust, pp. 30-35, July 2009.
[17] C. Yin and G. Qu, “Temperature-aware cooperative ring oscillatorPUF,” Hardware-Oriented Security and Trust, 2009. HOST ’09. IEEEInternational Workshop on, pp. 36-42, July 2009.
[18] R. Kumar, H.K. Chandrikakutty, and S. Kundu, “On improving relia-bility of delay based Physically Unclonable Functions under temperaturevariations,” IEEE International Symposium on Hardware-Oriented Secu-rity and Trust (HOST), pp. 142-147, June 2011.
[19] M. S. Kirkpatrick and E. Bertino, “Software techniques to combat driftin puf-based authentication systems,” in Workshop on Secure Componentand System Identification, 2010.
[20] A. Maiti, L. McDougall, P. Schaumont, “The Impact of Aging on anFPGA-Based Physical Unclonable Function,” in Field ProgrammableLogic and Applications (FPL), pp. 151-156, Sept. 2011.
[21] J. Abella , X. Vera and A. Gonzalez, “Penelope: The NBTI-awareprocessor”, Proc. Int. Symp. Microarchitect., pp. 85-96, 2007.
[22] A. Maiti et al., “A large scale characterization of RO-PUF,” in Proc.IEEE Int. Hardware-Oriented Security and Trust (HOST) Symp, pp. 94-99, 2010.
[23] A. Calimera, E. Macii, and M. Poncino, “NBTI-aware power gatingfor concurrent leakage and aging optimization,” In Proceedings of the14th ACM/IEEE international symposium on Low power electronics anddesign (ISLPED ’09), pp. 127-132, 2009.
[24] S. Krishnappa, H. Singh, and H. Mahmoodi, “Incorporating Effects ofProcess, Voltage, and Temperature Variation in BTI Model for CircuitDesign,” IEEE Latin American Symposium on Circuits and Systems,pp. 236-239, 2010.
[25] T. Ritter, Randomness tests: a literature survey.[Online]. Available:http://www.ciphersbyritter.com/RES/RANDTEST.HTM
[26] A. Calimera, E. Macii, and M. Poncino, “NBTI-aware sleep transistordesign for reliable power-gating,” in Proc. of GLSVLSI, pp. 333-338,2009.
[27] Synopsys, http://www.synopsys.com/
[28] Z. Qi and M.R. Stan, “NBTI resilient circuits using adaptive bodybiasing,” In Proc. GLSVLSI, ACM, pp. 285-290, 2008.
[29] L. Zhang and R.P. Dick, “Scheduled voltage scaling for increasinglifetime in the presence of NBTI,” In Proc. of ASP-DAC, pp. 492-497,2009.
[30] J.H. van Lint. Introduction to Coding Theory. Springer-Verlag, 1992.
[31] http://www.caimicro.com/web cn/upload/caimicro bch intro en v1.2.pdf
[32] T. Nigam, B. Parameshwaran, and G. Krause, “ Accurate productlifetime predictions based on device-level measurements,” In Proc. IRPS,pp. 634−639, 2009.