4 ms·

Two factors are important for a viable anti-tamper technology. First, the recipient must have an available technology to verify integrity. The random beads/ric

Two factors are important for a viable anti-tamper technology. First, the recipient must have an available technology to verify integrity. The random beads/rice method is cool because you just need a digital camera. Software that works like face recognition (a set of distance vectors between identifiable features) can run in seconds. There are 3D scanners that can snapshot an object to such accuracy that fingerprints will show up like mountain ranges. The point cloud for a laptop PCB is many gigabytes. We can send a hash to save space. Of course it's totally useless to an average person who doesn't have the same advanced scanner which costs millions. X-rays, which are super useful for supply chain integrity present a similar problem. But that creates a second problem, worse in a way. The technology cannot be too good. The scanner will pick up the thermal drift of solder during transit. That, or speck of dust, will then throw up a false positive. Unless you visually verify what the disturbance is (can't do that with the hash, need the full scan and it's time consuming) now you have a suspect device. Because digital devices might be maliciously soft-modified in undetectable ways, we can't take a chance. a $1000 device has to go in the trash. Too many false positives and it's not viable. My feeling is that several low-tech methods in cascade (to create defence in depth) are better than any single hi-tech method.

gcr 4y ago
Lossily compress the point cloud then, come on. For thin two-sided objects like laptops, rasterize the depth map of each side. You might be interested in how biometric hashes for things like fingerprints or irises work, since these have the same problems: the biometric is decomposed into a “stable” part and an “unstable” part. The stable part is hashed and the unstable part is encoded as a residual. W.J.Scheirer and T.E.Boult’s work on bipartite biotokens is one potential implementation
- nonrandomstring 4y ago
  > W.J.Scheirer and T.E.Boult's work on bipartite biotokens is one potential implementation Cheers. Something maybe like that already going on with a bespoke principal component analysis. But finding the needle in the haystack is knowing what _might_ change and be significant. Perhaps someone re-flashes a EEPROM, leaving only tiny dimple in a gold PCB pad. A system sensitive enough to pick that up doesn't necessarily know that a micro-fracture caused by vibration isn't a threat. What you're suggesting might be good for image processing of the plastic bead vacuum shield though.

gcr 4y ago

Lossily compress the point cloud then, come on. For thin two-sided objects like laptops, rasterize the depth map of each side. You might be interested in how biometric hashes for things like fingerprints or irises work, since these have the same problems: the biometric is decomposed into a “stable” part and an “unstable” part. The stable part is hashed and the unstable part is encoded as a residual. W.J.Scheirer and T.E.Boult’s work on bipartite biotokens is one potential implementation

nonrandomstring 4y ago
> W.J.Scheirer and T.E.Boult's work on bipartite biotokens is one potential implementation Cheers. Something maybe like that already going on with a bespoke principal component analysis. But finding the needle in the haystack is knowing what _might_ change and be significant. Perhaps someone re-flashes a EEPROM, leaving only tiny dimple in a gold PCB pad. A system sensitive enough to pick that up doesn't necessarily know that a micro-fracture caused by vibration isn't a threat. What you're suggesting might be good for image processing of the plastic bead vacuum shield though.