Physical analysis, trimming and editing of nanoscale IC function with backside FIB processing

On the Physical Security of Physically Unclonable Functions

“…Based on the already published results in the literature [14,74,34,64,35] and results of this work, it has become apparent that digital Intrinsic PUFs are not sensitive to sample preparation and optical semi-invasive attacks in their current configurations. By launching different classes of semi-invasive attacks, we demonstrated that Intrinsic PUFs are vulnerable to these attacks their responses can be predicted, manipulated and probed.…”

“…Therefore, an NVM might be replaced with a PUF to make the secrets volatile, harder to extract and even tamper-evident to invasive attacks. Indeed, one can deploy delay-based PUFs and distribute its elements on the whole area of a chip to develop a sensor to detect the FIB microprobing attempts by observing the intrinsic behavior of the PUF [14,74].…”

“…It has been shown that mechanical stress from depackaging and substrate thinning have negligible effects on the absolute and relative frequencies of ring-oscillators [14]. In another experiment, it has been reported that removing most of the bulk silicon, down to the bottom of the n-wells, does not alter the delays of the inverter chains [74]. Moreover, different successful semi-invasive attacks have been reported on silicon intrinsic PUF instances without affecting the challenge-response behavior of the PUFs [54,34,64].…”

“…If the attacker tries to measure the delays of each stage using invasive microprobing, the intrinsic internal delays of the PUF is changed with a high probability [74]. In other words, the challenge-response behavior of the PUF is altered, and therefore, the PUF will be destroyed.…”

“…To this end, it is possible to have an accurate delay map of the LEs on a second platform and try to utilize those, whose delays are close to the stages of the target PUF. Another option, though it is much more expensive, is to edit the circuit delays of the second platform with the help of FIB [74] to obtain timings close to timings of the target PUF. Thus, although achieving a physical clone of an Arbiter PUF is an onerous task, it is in principle possible.…”