![page 35
Quantum Tunneling PUFs .
“0” “1”
50% 50%
[Chuang, JSSC 2019]
[Wu, ISSCC 2018]
50% 50%
“0” “1”
▪ Only one tunneling path will be generated in two of the NMOS transistors
▪ Reading out the tunneling current of PUF cells → deriving PUF bits](https://image.slidesharecdn.com/puflecture2-230606124001-5c4a0740/85/PUF_lecture2-pdf-35-320.jpg)
PUF_lecture2.pdf
- 2. How to make PUFs Reliable & Why it is important ? Dr. Kent Chuang 2022 September
- 3. Outline . 1. Recap 2. Why reliability is important? 3. Methods to improve reliability 4. Intrinsically reliable PUF using quantum tunneling 5. Conclusion and Outlook
- 4. Outline . 2. Why reliability is important? 3. Methods to improve reliability 4. Intrinsically reliable PUF using quantum tunneling 5. Conclusion and Outlook
- 5. page 5 Recap . In the previous lecture, we have introduced ▪ The importance of hardware security ▪ Why we need root of trust and PUF ▪ What is a PUF? ▪ What are the important PUF properties?
- 6. page 6 A PUF consists of the following properties: ▪ Physical unclonability ▪ Evaluability ▪ Uniqueness ▪ Reliability Inborn unique secret of a chip PUF is a circuit with required properties .
- 7. page 7 ▪ PUF needs to be unique (having the “uniqueness” property) Secret key Integrated Circuit PUF Cryptographic Function Secure Communication Key generation K1 K2 K3 Let each chip generate its own key .
- 8. page 8 Key Generation Using Weak PUF . KDF Device Secret Auxiliary Input (Optional) Secret Key PUF Array 0 1 0 1 0 1 0 0 0 Readout Interface ▪ Unique device secret can be derived from the PUF array ▪ Secret key can be further derived by sending the device secret into the key derivation function (KDF)
- 9. Outline . 1. Recap 3. Methods to improve reliability 4. Intrinsically reliable PUF using quantum tunneling 5. Conclusion and Outlook
- 10. page 10 0110010… 0110110… 0010110… t Bit-errors T1 T2 Tn → Target: no bit error at any time and operating conditions K1 Kn≠K1 Potential issue: the resulting keys may be different K2 ≠K1 The same PUF queried at different time Reliability: consistency of PUF responses .
- 11. page 11 Crypto functions cannot tolerate errors . ▪ For security reason, we want: – A similar key should result in very different message/cipher pair ▪ In this case, if a key is corrupted – Almost impossible to recover the corrupted messages M C=Enc(M, K) Encryption K M’=Dec(C, K’) Decryption K’ HD(K, K’)=1 FHD(M, M’) ≈ 0.5
- 12. page 12 Risks of Losing Information . Core Encryption Flash Decryption Data Encrypted Data Chip/Device ▪ Sensitive data stored in Flash needs to be encrypted for security ▪ Decrypted data will be corrupted if the key has errors Example: Flash Encryption PUF
- 13. page 13 Failing Authentications . KDF Auxiliary Input PUF Public Key Private Key PUF-based Key-Pair Server Sign Enroll Signature Challenge ▪ Server authenticate the device by verifying its signature ▪ Signature will not match the public key if the private key has errors → Authentication Failed Chip/Device
- 14. page 14 Reliability may be very different . “0” “1” “1” “0” SRAM PUF Quantum Tunneling PUF ▪ Power up results may be different each time → Poor reliability 50% 50% “0” “1” ▪ Tunneling path remains the same each readout → Ideal reliability
- 15. page 15 SRAM PUF . ▪ 2D array of 1-bit memory cells ▪ Using the mismatch between the cross-coupled inverters ord ine bit ine bit ine 6T-SRAM cell I1 I2 “1” “0” “0” “1” Bi-stable states I1 I2 I2 I1 Two possible outcomes after power-up
- 16. page 16 SRAM PUF cells may be unreliable . ▪ Mismatches are random → It is possible to have very small mismatches ▪ An SRAM can enter noise-sensitive metastable state → SRAM PUF data may change in different power-ups ord ine bit ine bit ine identical identical 50% “0” “1” 50% “1” “0” Bit errors!
- 17. page 17 Deriving secret keys from PUF data . ▪ Post-processing: including error-correction function – Robust mathematical algorithms, e.g. BCH – Need memory and complex logics → resource consuming ▪ Stabilization: lightweight methods to reduce native error – Temporal majority voting, dark-bit masking, burn-in, … Readout Interface Post- processing n-bit k-bit Stabilization m-bit
- 18. Outline . 1. Recap 2. Why reliability is important? 4. Intrinsically reliable PUF using quantum tunneling 5. Conclusion and Outlook
- 19. page 19 Error Correction Codes . Generator Data (Data || Parity) Decoder Data Error ▪ Parity bits are generated based on the data and the ECC algorithm ▪ Data corrupted by a limited number of errors can be decoded back ▪ More parity bits → better error tolerance – BCH(255, 247, 1) → Can correct maximum 1-bit error with 8 parity bits – BCH(255, 199, 7) → Can correct maximum 7-bit error with 56 parity bits
- 20. page 20 Deriving a PUF-based key . SRAM PUF bits PUF PUF bits with error R.N. Parity PUF (R.N. + Parity) PUF (R.N.+ Parity) PUF’ XOR NVM 255-bit BCH-encoded Helper data Enrollment Derivation 255-bit XOR Parity’ R.N.’ With error PUF Key derivation Parity R.N. PUF (R.N.+ Parity) XOR Error corrected Helper data
- 21. page 21 Reconstructing a PUF-masked key . SRAM PUF bits PUF PUF bits with error Key Parity 255-bit PUF (Key + Parity) PUF (Key + Parity) PUF’ XOR XOR Parity’ Key’ Key NVM 255-bit BCH-encoded key 255-bit Helper data BCH decoding Key and parity with errors Enrollment Reconstruction
- 22. page 22 Post-processing is not an ideal solution . ▪ Need random number and NVM for both methods ▪ No advantage in terms of cost – Comparing to storing keys in NVM ▪ It does provide better physical security – Helper data can be a public information – But it cannot be modified → access control required for NVM (cost) PUF Helper data NVM Post- processing RNG
- 23. page 23 Lightweight methods for error reduction . ▪ Typically applied before post-processing – To reduce the required error-correction strength – Some methods may replace error-correction ▪ Temporal majority voting, dark-bit masking, … Readout Interface Post- processing n-bit k-bit Stabilization m-bit
- 24. page 24 Reduce errors by majority voting . ▪ More errors are averaged out if more readouts are performed ▪ Very inefficient if the native error rate is high N Error rate
- 25. page 25 Screen out PUF bits that cause errors . ▪ Distinguishing unstable bits is challenging and time-consuming ▪ The mask information has to be stored – NVM is needed How to generate and store the mask?
- 26. page 26 Combining these methods . Post- processing (ECC) TMV Masking Mask (NVM) Initial Test Read Enroll Helper Data (NVM) Enroll Read Read n-bit m-bit m-bit k-bit ▪ Enrollment: find unstable cells and write helper data ▪ In-Field: stabilize PUF data and derive the secret key
- 27. page 27 PUFs need to be intrinsically reliable . Readout Interface Post- processing n-bit k-bit Stabilization m-bit ▪ Error correction is too costly – Extra cost on computation resources, storage and latency ▪ Stabilization techniques are insufficient Costly Insufficient
- 28. page 28 Having highly reliable PUF is beneficial . ▪ No error-correction, no stabilization, no NVM ▪ Instant ready PUF-based key Highly reliable PUF-based secret key Why most PUFs are not intrinsically reliable? Readout Interface Entropy Extraction (optional) n-bit k-bit
- 29. page 29 Variations in a PUF can be too small . ▪ Variations must be “active y” enhanced/created → exploiting time-dependent variability ord ine bit ine bit ine VT=0.31V VT=0.3V 60% 40% “1” “0” “0” “1” Prone to transient fluctuation and aging Can we increase this VT to 0.4V?
- 30. page 30 Increase mismatches through burn-in . VT Percentage Burn-in VT Percentage For example: ▪ VT of the two transistors originally follow the same distribution ▪ The distribution can be separated by applying burn-in mechanism
- 31. page 31 SRAM PUF enhanced by BTI effects . ▪ Program the SRAMs oppose to the power-up state – Mismatch increased due to BTI stress ▪ Time consuming and partially recoverable ord ine bit ine bit ine weak nBTI Becomes weaker → Less difference 0 VDD power-up ord ine bit ine bit ine 0 VDD program weak nBTI Even weaker → More difference Don’t keep the power-up pattern *BTI: biased temperature instability R. Maes, “Countering-the-effects-of-silicon-aging-on-SRAM-PUFs,” Symp. HOST 2014
- 32. page 32 • BER=0% is reached for 125s stress → effective but takes too long • Resulting HD~0.47 → uniqueness is affected by peripheral circuits SA PUF enhanced by hot-carrier injection . mismatch M. Bhargava, et. a , “A high re iabi ity PUF using hot carrier injection based response reinforcement,” CHES 2013
- 33. Outline . 1. Recap 2. Why reliability is important? 3. Methods to improve reliability 5. Conclusion and Outlook
- 34. page 34 + - V stress Vstress 0 0 Stress Irreversible→ reliable Metal Gate Substrate Traps Oxide Metal Gate Substrate + - V stress Percolation path Substrate Metal Gate Wearout Soft Oxide Breakdown (Tunneling) V stress + - Highly reliable PUF using oxide tunneling . ▪ MOSFET devices have an insulating gate oxide layer ▪ Traps are generated by voltage stress, eventually form a tunneling path ▪ Tunneling path will not vanish after being generated
- 35. page 35 Quantum Tunneling PUFs . “0” “1” 50% 50% [Chuang, JSSC 2019] [Wu, ISSCC 2018] 50% 50% “0” “1” ▪ Only one tunneling path will be generated in two of the NMOS transistors ▪ Reading out the tunneling current of PUF cells → deriving PUF bits
- 36. page 36 The self-limiting mechanism . ▪ Current and voltage are limited by the PMOS selector ▪ Ensuring only one BD (tunneling) spot in a PUF cell Δ = Vstress - VDS Vstress VG VDS IBD Reduced stress voltage → No breakdown Limited BD current → Only soft-BD Define saturation current (current limit) Vstress VG Apply constant voltage stress Time to breakdown (tBD) Chuang, et. al, A Physically Unclonable Function Using Soft Oxide Breakdown Featuring 0% Native BER and 51.8fJ/bit in 40nm CMOS, JSSC 2019
- 37. Outline . 1. Recap 2. Why reliability is important? 3. Methods to improve reliability 4. Intrinsically reliable PUF using quantum tunneling
- 38. PUFsecurity page 38 page 38 Conclusion . To ensure the correctness of PUF-based security app ications … ▪ Reliability of PUFs is with high importance ▪ Conventional reliability improvement methods are insufficient ▪ Highly reliable Quantum Tunneling PUF is introduced … brings up reliable and efficient security solutions
- 39. PUFsecurity page 39 page 39 Outlook . Coming up: ▪ Popular circuit implementation of PUFs ▪ Detailed design and analysis of quantum tunneling PUFs ▪ Benchmark of popular PUF implementations ▪ Examples of PUF-based security applications
- 40. Thank you! More educational materials? Feel free to follow us!