A Perspective on Security and Trust Requirements for the Future Dr. - PowerPoint PPT Presentation

A Perspective on Security and Trust Requirements for the Future Dr. Kenneth Plaks International Symposium on Physical Design April 14-17, 2019 Distribution Statement A: Approved for Public Release, Distribution Unlimited

Hardware Security in the Field US Air Force Distribution Statement A: Approved for Public Release, Distribution Unlimited

The importance of electronics US Air Force Distribution Statement A: Approved for Public Release, Distribution Unlimited

Commercial and Military have similar needs… Commercial Military IP protection and overproduction Export leakage – ITAR concerns • • License Enforcement Anti Tamper • • Brand identity and dependability Trust – reliability and no malicious insertions • • Distribution Statement A: Approved for Public Release, Distribution Unlimited

But we place a much higher priority on security Security Area Vs Power Delay Power Delay Area Commercial Optimization Defense Optimization What would physical design look like if we optimized for security, instead of area? Distribution Statement A: Approved for Public Release, Distribution Unlimited

(U) Physical design and security Side Channels Sensors Digital World Real World Physical Design Distribution Statement A: Approved for Public Release, Distribution Unlimited

Sensors For this talk, sensors are things that provide data from the real world to the digital world • Military uses sensors to determine where a chip has been • Military uses sensors as a root of trust • We will use the DARPA SHIELD project as an example • Image courtesy of SRI International Image courtesy of Northrop Grumman Distribution Statement A: Approved for Public Release, Distribution Unlimited

Sensor Example: DARPA SHIELD 100 µm Charge 31 µm Pump One-Time PUF & Key SHIELD Specifications Programmable Voltage Sensor Memory Clamp Unique Key Storage • 31 µm Full 256-bit AES encryption engine • Analog 50 µm 57 µm Unpowered, passive intrusion sensors • Modulator 100 µm RF power and communication • 21 µm 58 µm Regulator Rectifier Transfer fragility • 100µm x 100µm Common Bias Modulator • Digital 50 µW Total Power • POR ADC Operating temp < 120  C • Osc 42 µm 38 µm Cost < $0.01 per dielet • RNG 70 µm 22 µm Osc Dielet floorplan (Northrop Grumman) Prototype dielet layout (SRI) 14nm CMOS 28nm CMOS Asymmetric Security Non- resettable, “always on” intrusion sensors on dielet • On- board encryption symmetric key that cannot be “coaxed” from dielet • ID and Key are unique to the individual host IC (not just the part number) • Interrogation history (date, time, location) stored on secure server • Built-in fragility structures kill dielet if removal from host is attempted • SHIELD makes counterfeiting too expensive and too hard to do. Images courtesy of Northrop Grumman and SRI International Distribution Statement A: Approved for Public Release, Distribution Unlimited

SHIELD – Xray and RF Sensor Testing Sensor Radiation Sensitivity 3 2.75 2.5 Voltage (V) 2.25 2 1.75 1.5 0 1 2 3 4 5 6 Dose in KRad(Si) Mini-x- ray test fixture in Draper’s Radiation Effects Lab Stock photos of anechoic chamber, antenna & probe. RF testing was carried out in a secure lab at Draper Distribution Statement A: Approved for Public Release, Distribution Unlimited

Physical Vulnerabilities of PUFs The 1 st silicon iteration of a DoD PUF failed due to its output voltages being severely skewed in the negative direction (toward 0V) The root cause of the voltage skew was the layout proximity effect which is a dominant effect in nanoscale devices SHIELD, DARPA The PUF voltages should have been uniformly distributed; however, testing revealed that most of the voltages were skewed negatively toward 0V. Distribution Statement A: Approved for Public Release, Distribution Unlimited

Side channels For this talk, side channels are ways of getting data from the digital world to the real world • Not talking about SPECTER and MELTDOWN • Military uses side channels to find malicious circuits • How can we interrogate a circuit for malice, when we don’t trust the circuit in the first place? • What aspects of physical design could enhance security? • Distribution Statement A: Approved for Public Release, Distribution Unlimited

The trojan challenge in three charts In this region In this region customization and programmability and optimization flexibility dominate dominate Sources: IBS; A. Olofsson, “Silicon Compilers - Version 2.0”, keynote, Proc. ISPD, 2018 Moore’s law makes SOC’s possible Distribution Statement A: Approved for Public Release, Distribution Unlimited

Technical Observation But we have lost herd immunity Distribution Statement A: Approved for Public Release, Distribution Unlimited

Hardware Trojans Moore’s law also makes defense even harder Distribution Statement A: Approved for Public Release, Distribution Unlimited

One possible approach: physical timing side channels Trojan impact on timing should be observable even without activating the trojan • But voltage variation makes that hard to measure • H. Kaul, M. Anders, S. Hsu, A. Agarwal, R. Krishnamurthy and S. Borkar, "Near-threshold voltage (NTV) design — Opportunities and challenges," DAC Design Automation Conference 2012 , San Francisco, CA, 2012, pp. 1149-1154. Distribution Statement A: Approved for Public Release, Distribution Unlimited

Solution to Complications: Voltage Noise • Process Drift • Count A. Vakil, H. Homayoun, and A. Sasan, Proceedings of the 24th Asia and Expected Path Delay South Pacific Design Automation Conference. ACM, 2019, pp. 152– 159. Avesta Sasan, GMU Path Delay IR Annotated Timing Analysis NN Process Watchdog 1 2 3 4 5 6 7 https://dl.acm.org/citation.cfm?id=3287683 Distribution Statement A: Approved for Public Release, Distribution Unlimited

AVATAR (IR-ATA): Annotating the Timing Impact of Voltage drop and Noise Spice Verification: the improved accuracy STA Improvement: Impact of using AVATAR(IR- of AVATAR (IR-ATA) in capturing the timing ATA) for reporting the timing slack: impact of Voltage drop (STA versus Spice) The released slack could be used for PPA • improvement. Avesta Sasan, GMU Distribution Statement A: Approved for Public Release, Distribution Unlimited

PPA Improvement Power reduction: The released slack (from Performance boost: The released slack (from using AVATAR for voltage drop and voltage using AVATAR for voltage drop and voltage noise noise modeling) is used for ECOs targeting the modeling) in critical timing paths, allow the reduction of leakage and dynamic power. physical designer to shorten the clock cycle time, leading to a higher performance design. Reduces Dynamic power • Reduces Leakage power • Increases max frequency • Reduces area • Avesta Sasan, GMU Distribution Statement A: Approved for Public Release, Distribution Unlimited

Trojan Detection Rate The new voltage variation aware timing model (GTM) along with NN process watch dog Small Cert Medium HC can significantly improve the chances of Trojan detection without having access to a Large LC Golden IC. Ethernet S38417 AES128 100 100 90 80 80 80 Percentage Percentage Percentage 70 Detection Detection Detection 60 60 60 50 40 40 40 30 20 20 20 10 0 0 0 Timing Model Used Timing Model Used Timing Model Used TT Trojans Detection Rate TT Trojans Detection Rate TT Trojans Detection Rate 100 100 100 80 Percentage 80 Percentage 80 Percentage Detection Detection Detection 60 60 60 40 40 40 20 20 20 0 0 0 Timing Model Used Timing Model Used Timing Model Used TP Trojans Detection Rate TP Trojans Detection Rate TP Trojans Detection Rate Trojan Detection Design and test Flow Avesta Sasan, GMU Distribution Statement A: Approved for Public Release, Distribution Unlimited

Security aware physical design Routing VIA: Adapt via size to help enhance Decap: Insert decaps to help stabilize power estimate power estimation accuracy Avesta Sasan, GMU Leakage reduction Upsize critical VIAs • • Adds less routing violations Lower space overhead • • Distribution Statement A: Approved for Public Release, Distribution Unlimited

Conclusion Physical design is where the digital becomes real • Wikimedia Commons There are several opportunities to enhance security • Or inadvertently break it… • By adding security considerations to physical design • offers the opportunity to make better chips, with lower security risk To make security-aware physical design a reality Distribution Statement A: Approved for Public Release, Distribution Unlimited

www.darpa.mil Distribution Statement A: Approved for Public Release, Distribution Unlimited