The security characteristics that may be leveraged have been summarized by AFRL researchers in published work3 and are listed below.
- Non-volatility: Memristors retain their memristance value even when the power is turned OFF.
- Bi-directionality: Some bipolar memristors exhibit similar current-voltage characteristics irrespective of the polarity of the applied voltage or current.
- Non-linearity: The I-V characteristics of memristors are highly non-linear due to their time-dependent behavior. Also, the High Resistance State (HRS) to Low Resistance State (LRS) ratio is typically on the order of 103-106.
- Formation process: For many memristors, a separate forming step (Vf) is required to initialize the memristor to the LRS. Prior to this point, the memristor behaves as a linear resistor.
- Memristance drift: On applying an input voltage (positive or negative) across certain metal-oxide memristors, the memristance changes because of the movement of dopants, a process called memristance drift. The amount of drift depends on the polarity, amplitude, and duration of the applied voltage.
- Process variations: The memristance of a memristor is affected by process-variation induced changes in its dimensions and dopant concentration. Furthermore, the effects of variation in the thickness of the memristor upon its memristance values are highly non-linear (more significantly for the LRS than the HRS).
- Radiation-hardness: Some memristor devices are inherently radiation-hard due to their material properties.
- Temperature stability: The LRS and HRS values are highly stable in the case of a TiO2 memristor since the temperature coefficient of resistance for TiO2 is very small (less than -3.82×10-3/K). However, the switching speed of the memristor varies with temperature because of the change in dopant atom mobility.
All of these characteristics with the exception of nonvolatility and radiation-hardness pose problems when designing memory and logic circuits using a metal-oxide memristor, but can be useful in the context of security. It is for these reasons this work will identify, from the myriad of choices in materials, the most suitable material stack for focusing future memristor device research and technologies. This will be accomplished by fabricating memristor devices with differing material stacks and comparing their memristance performance, both endurance and resistance drift, when subjected to variable temperature operational conditions. Additionally, room temperature comparisons of the effects of total ionizing dose on device performance will also be assessed.
A complementary research initiative that will leverage the results of the memristor research described above will result in a more fundamental understanding of the requirements for designing a robust PUF system.
A PUF can be described as a fingerprint that can be used to uniquely identify individual integrated circuits (ICs). PUFs are unique in that no two devices will have the same signature and are unclonable due to the inherent infeasibility required to create two devices with the same signature. In the literature, both uniqueness and unclonability have been attributed to intrinsic variations resulting from nonuniform manufacturing process. There is variability in the complex physical processes associated with IC design and manufacturing. This creates a natural defense to an attacker whom now must either control the noise, or selectively and predictively change manufacturing parameters without disrupting the functional correctness of the resulting ICs. This portion of research will seek to gain a better understanding of whether these fundamental assumptions are valid, and formalize standards that define what makes a suitable PUF.