State-of-the-art Machine Learning based Modeling Attacks Phuong Ha - PowerPoint PPT Presentation

The Interpose PUF (iPUF): Secure PUF Design against State-of-the-art Machine Learning based Modeling Attacks Phuong Ha Nguyen, Durga P. Sahoo, Kaleel Mahmood, Chenglu Jin, Ulrich Rührmair and Marten van Dijk Secure Computation Laboratory Department of Electrical & Computer Engineering University of Connecticut Ha Durga Chenglu Kaleel Marten Uli CHES 2019

Content 1. Concept - Overview - Motivation 2. Strong PUFs: APUF, XOR APUF and Interpose PUF (iPUF) 3. Short-term Reliability 4. Reliability based modeling attacks on XOR PUF: understanding 5. Interpose PUF – a lightweight PUF which is secure against state-of-the art modeling attacks 6. Conclusion 2

1 . Concept - Overview - Motivation 3

Concept - Overview – Motivation [1] Hardware Challenge C Response R Primitive [Device] PUF’s Category : Nature: process variation – physically unclonability - unique Application: device Identification, authentication Weak PUF - small #CRPs: and crypto key generation RO PUF, SRAM PUF, etc. Strong PUF – large #CRPs: No Security Proof: Security Proof: Broken but lightweight: APUF, XOR APUF, Feed Power Grid PUF, LPN PUFs - heavy Forward PUF, Clock PUF, Crossbard PUF Lightweight Secure PUF, Bistable Ring PUF, MPUF etc. 4

Concept - Overview – Motivation [2] Hardware Challenge C Response R Primitive [Device] PUF’s Category : Nature: process variation – physically unclonability - unique Application: device Identification, authentication Weak PUF - small #CRPs: and crypto key generation RO PUF, SRAM PUF, etc. Strong PUF – large #CRPs: No Security Proof: Security Proof: Broken but lightweight: APUF, XOR APUF, Feed Power Grid PUF, LPN PUFs - heavy Forward PUF, Clock PUF, Crossbard PUF Lightweight Secure PUF, Bistable Ring PUF, MPUF etc. 5

Concept - Overview – Motivation [3] Hardware Challenge C Response R Primitive [Device] PUF’s Modeling Attacks on CRPs only : PUF’s Category : Weak PUF - small #CRPs: Advanced ML attacks Classical ML attacks RO PUF, SRAM PUF, etc. – noisy CRPs: – reliable CRPs: CMA-ES + noisy CRPs Strong PUF – large #CRPs: Support Vector Machine (SVM), Logistic Regression (LR), Evolution Strategy (ES), No Security Proof: Security Proof: Broken but lightweight: Covariance Matrix Adaptation APUF, XOR APUF, Feed Power Grid PUF, LPN PUFs ES (CMA-ES), Perceptron, - heavy Forward PUF, Clock PUF, Boolean Attacks, Deep Neural Crossbard PUF Lightweight Secure PUF, Network Attacks (DNN) Bistable Ring PUF, MPUF etc. 6

Concept - Overview – Motivation [4] Hardware Challenge C Response R Primitive [Device] PUF’s Modeling Attacks with CRPs only : PUF’s Category : Weak PUF - small #CRPs: Advanced ML attacks Classical ML attacks RO PUF, SRAM PUF, etc. – noisy CRPs: – reliable CRPs: CMA-ES + noisy CRPs Strong PUF – large #CRPs: Support Vector Machine (SVM), Logistic Regression (LR), Evolution Strategy (ES), No Security Proof: Security Proof: Broken but lightweight: Covariance Matrix Adaptation APUF, XOR APUF, Feed Power Grid PUF, LPN PUFs ES (CMA-ES), Perceptron, - heavy Forward PUF, Clock PUF, Boolean Attacks, Deep Neural Crossbard PUF Lightweight Secure PUF, Network Attacks (DNN) Bistable Ring PUF, MPUF etc. 7

Concept - Overview – Motivation [5] Hardware Challenge C Response R Primitive [Device] PUF’s Modeling Attacks with CRPs only : PUF’s Category : Weak PUF - small #CRPs: Advanced ML attacks Classical ML attacks RO PUF, SRAM PUF, etc. – noisy CRPs: – reliable CRPs: CMA-ES + noisy CRPs Strong PUF – large #CRPs: Support Vector Machine (SVM), Logistic Regression (LR), Evolution Strategy (ES), No Security Proof: Security Proof: Broken but lightweight: Covariance Matrix Adaptation Arbiter PUF/APUF, XOR Power Grid PUF, LPN PUFs ES (CMA-ES), Perceptron, - Large HW APUF, Feed Forward Clock PUF, Boolean Attacks, Deep Neural Crossbar PUF footprint PUF, Lightweight Secure Network Attacks (DNN) PUF, Bistable Ring PUF. 8

Concept - Overview – Motivation [6] Hardware Challenge C Response R Primitive [Device] PUF’s Modeling Attacks with CRPs only : PUF’s Category : Weak PUF - small #CRPs: Advanced ML attacks Classical ML attacks RO PUF, SRAM PUF, etc. – noisy CRPs: – reliable CRPs: CMA-ES + noisy CRPs Strong PUF – large #CRPs: Security Proof Broken but lightweight: Vulnerability Arbiter PUF/APUF, XOR Lightweight, APUF, Feed Forward Precise Math. Model PUF, Lightweight Secure XOR APUF PUF, Bistable Ring PUF. 9

Concept - Overview – Motivation [7] Hardware Challenge C Response R Primitive [Device] PUF’s Modeling Attacks with CRPs only : PUF’s Category : Weak PUF - small #CRPs: Advanced ML attacks Classical ML attacks RO PUF, SRAM PUF, etc. – noisy CRPs: – reliable CRPs: CMA-ES + noisy CRPs Strong PUF – large #CRPs: Security Proof Broken but lightweight: Security Proof Arbiter PUF/APUF, XOR Lightweight, APUF, Feed Forward Precise Math. Model PUF, Lightweight Secure XOR APUF interpose PUF (iPUF) PUF, Bistable Ring PUF. 10

Concept - Overview – Motivation [8] Hardware Challenge C Response R Primitive [Device] PUF’s Modeling Attacks with CRPs only : PUF’s Category : Weak PUF - small #CRPs: Advanced ML attacks Classical ML attacks RO PUF, SRAM PUF, etc. – noisy CRPs: – reliable CRPs: CMA-ES + noisy CRPs Strong PUF – large #CRPs: Security Proof Broken but lightweight: Security Proof Security Philosophy Arbiter PUF/APUF, XOR Lightweight, APUF, Feed Forward Precise Math. Model PUF, Lightweight Secure XOR APUF interpose PUF (iPUF) PUF, Bistable Ring PUF. Design Philosophy 11

2. APUF- XOR APUF -iPUF 12

APUF, XOR APUF and iPUF [1] Arbiter PUF (APUF) [1] x-XOR APUF Interpose PUF (iPUF) - Extremely lightweight and large number of CRPs i.e, 2 𝑜 CRPs - Environmental noises make the PUF’s outputs unreliable sometimes - Not secure against modeling attacks

APUF, XOR APUF and iPUF [2] Arbiter PUF (APUF) x-XOR APUF

APUF, XOR APUF and iPUF [3] The Interpose PUF / iPUF 15

APUF, XOR APUF and iPUF [4] Interpose PUF (iPUF) x-XOR Arbiter PUF Arbiter PUF (APUF) 𝑦, 𝑧 − 𝐽𝑄𝑉𝐺 ≈ 𝑧 + 𝑦 2 − 𝑌𝑃𝑆 𝑄𝑉𝐺 if a is inserted at the middle Precise non-linear model + CRPs + classical ML = impractically softwarelly clonable • Δ > 0 → 𝑠 = 1. 𝑃𝑢ℎ𝑓𝑠𝑥𝑗𝑡𝑓 𝑠 = 0 • Δ = 𝒆 𝒗𝒒𝒒𝒇𝒔 − 𝒆 𝒎𝒑𝒙𝒇𝒔 = 𝒙 ⋅ 𝚾 Precise non-linear model + CRPs + classical ML • 𝒙 ∶ 𝑣𝑜𝑗𝑟𝑣𝑓 𝑥𝑓𝑗𝑕ℎ𝑢 𝑤𝑓𝑑𝑢𝑝𝑠, = impractically softwarelly clonable 𝑒𝑓𝑚𝑏𝑧 𝑠𝑓𝑞𝑠𝑓𝑡𝑓𝑜𝑢𝑏𝑢𝑗𝑝𝑜 𝑔𝑝𝑠 𝑏𝑜𝑧 𝐵𝑄𝑉𝐺 𝑗𝑜𝑡𝑢𝑏𝑜𝑑𝑓 • XOR APUF is not Secure against noisy CRPs + 𝚾 𝑗𝑡 𝑢ℎ𝑓 𝑞𝑏𝑠𝑗𝑢𝑧 𝑤𝑓𝑑𝑢𝑝𝑠 𝚾 𝑗 = 𝑘=𝑗,…,𝑜−1 1 − 𝒅 𝑘 , 𝑗 = 0, … , 𝑜 − 1 , 𝚾 𝑜 = 1 CMA-ES [Advanced ML]! (CHES2015) why? Precise linear model + CRPs + ML Why not for IPUF? 16 = practically and softwarelly clonable

APUF, XOR APUF and iPUF [5] Interpose PUF (iPUF) x-XOR Arbiter PUF Arbiter PUF (APUF) 𝑦, 𝑧 − 𝐽𝑄𝑉𝐺 ≈ 𝑧 + 𝑦 2 − 𝑌𝑃𝑆 𝑄𝑉𝐺 if a is inserted at the middle • Δ > 0 → 𝑠 = 1. 𝑃𝑢ℎ𝑓𝑠𝑥𝑗𝑡𝑓 𝑠 = 0 Precise non-linear model + CRPs + classical ML • Δ = 𝒆 𝒗𝒒𝒒𝒇𝒔 − 𝒆 𝒎𝒑𝒙𝒇𝒔 = 𝒙 ⋅ 𝚾 = impractically softwarelly clonable • 𝒙 ∶ 𝑣𝑜𝑗𝑟𝑣𝑓 𝑔𝑝𝑠 𝑏𝑜𝑧 𝐵𝑄𝑉𝐺 𝑗𝑜𝑡𝑢𝑏𝑜𝑑𝑓 Precise non-linear model + CRPs + classical ML • 𝚾 𝑗𝑡 𝑢ℎ𝑓 𝑞𝑏𝑠𝑗𝑢𝑧 𝑤𝑓𝑑𝑢𝑝𝑠 = impractically softwarelly clonable 𝚾 𝑗 = 𝑘=𝑗,…,𝑜−1 1 − 𝒅 𝑘 , 𝑗 = 0, … , 𝑜 − 1 , 𝚾 𝑜 = 1 XOR APUF is not Secure against noisy CRPs + • Precise linear model CMA-ES [Advanced ML]! (CHES2015) why? • Large CRP space • Why not for IPUF? Vulnerable to ML attacks 17

APUF, XOR APUF and iPUF [6] Interpose PUF (iPUF) x-XOR Arbiter PUF Arbiter PUF (APUF) 𝑦, 𝑧 − 𝐽𝑄𝑉𝐺 [2] ≈ 𝑧 + 𝑦 2 − 𝑌𝑃𝑆 𝑄𝑉𝐺 if a is inserted at the middle • • Δ > 0 → 𝑠 = 1. 𝑃𝑢ℎ𝑓𝑠𝑥𝑗𝑡𝑓 𝑠 = 0 Precise non-linear model • • Δ = 𝒆 𝒗𝒒𝒒𝒇𝒔 − 𝒆 𝒎𝒑𝒙𝒇𝒔 = 𝒙 ⋅ 𝚾 Large CRP space • • Secure against classical ML 𝒙 ∶ 𝑣𝑜𝑗𝑟𝑣𝑓 𝑔𝑝𝑠 𝑏𝑜𝑧 𝐵𝑄𝑉𝐺 𝑗𝑜𝑡𝑢𝑏𝑜𝑑𝑓 Precise non-linear model + CRPs + classical ML • • Vulnerable to advanced ML 𝚾 𝑗𝑡 𝑢ℎ𝑓 𝑞𝑏𝑠𝑗𝑢𝑧 𝑤𝑓𝑑𝑢𝑝𝑠 = impractically softwarelly clonable 𝚾 𝑗 = 𝑘=𝑗,…,𝑜−1 1 − 𝒅 𝑘 , 𝑗 = 0, … , 𝑜 − 1 , 𝚾 𝑜 = 1 XOR APUF is not Secure against noisy CRPs + • Precise linear model CMA-ES [Advanced ML]! (CHES2015) why? • Large CRP space • Why not for IPUF? Vulnerable to ML attacks 18