Introduction to PUFs

Though Question

• You're planning on leaving your laptop in a hotel room
• You suspect that criminals may sneak in, open your laptop, clone your hard-drive and plant a data skimming device
• How would you know if someone's been in your laptop?

How could you know if someone tampers with the device?

• Tamper seal? It needs something that is unique so that it is hard to clone or replace.
• It should also be inexpensive

One Solution

• Glitter nail polish
  • Paint over screws in laptop case
• Why is it good?
  • Randomness of glitter is very hard (if not impossible) to replicate

PUFs

• Physically unclonable function
• Concept of unique objects applied to the digital domain

Building a PUF

• Need something that varies randomly between chips
  • But doesn't change too much as a chip ages
• Need randomness that's difficult/cost-prohibitive to reproduce
• Want something cheap and easy to add to a system
  • Can't significantly affect performance, power, chip area, or reliability
  • Must be compatible with commercial fab processes
• Need lots of ways to read/test/challenge the system to prevent spoofing

IC-to-IC Entropy

• Process variations mean that every chip is different
• Sources of PUF variation
  • Digital features sizes are smaller than the wavelength of light used to print them
  • Some impurities exist in the substrates and dopants
  • Tolerance in mask alignments between processing stages
• This means
  • No two wires are exactly the same
  • No two transistors are exactly the same
  • Delays and initialization states vary

Three types of PUFs

• Memory
  • Look at what values memory cells naturally take on power-up
• Ring Oscillator
  • Design several identical ring oscillators
  • Count how many times each oscillates over a period
• Arbiter PUF (shown below)

Building a PUF: SRAM Example

• Power on each chip
• Read the initial values from local SDRAMs
• Figure out which SRAM cells always initialize the same way
• Use these values as your PUF responses

Caution: Many sources of Entropy

• Systematic process variations across chip wafers
• Randomness from temperature variation
• Randomness from hysteresis
  • Charges may persist between power cycles
  • Transistors (and interconnect) will degrade randomly with use, eventually flipping some initialization states

Difficulties in PUF Design

• Differentiate per device random variation from thermal, system, and hysteresis dependent
• COmpute consistent(ish) responses for different inputs across a wide range of operating conditions
  • Use error correcting codes to lose the "ish"
  • That reduces numbe of possible challenge response pairs, though...