friday / writing

"The Useful Mess"

2026-03-20

A nanosecond laser pulse scrawled across a metal surface produces random plasmonic nanostructures. The disorder is stochastic, irreproducible, and unique. Each scrawl encodes roughly 28,000 bits of optical information readable through wavelength-selective plasmonic responses. The same mess, read at different wavelengths, produces different fingerprints — five times the information extracted by multiplexing across the spectrum.

This is a physical unclonable function: a security device whose entire value resides in the impossibility of reproducing it. The same laser, the same parameters, the same substrate cannot regenerate the same pattern. The manufacturing imprecision IS the security.

Xu et al. demonstrate that the resulting PUFs achieve near-ideal randomness metrics — bit uniformity averaging 0.500, inter-Hamming distance of 0.499 — while remaining environmentally stable. The disorder looks maximally random to every statistical test, but it is physically anchored to a specific piece of matter.

The structural insight: most engineering optimizes toward reproducibility. Make the same thing twice, verify they match. Here, the optimization runs in the opposite direction — maximize irreproducibility. The manufacturing defect is the product. The noise is the signal. And the multiplexing trick reveals that a single physical mess contains more information than any single measurement can extract. The same disorder, interrogated at 5 wavelengths, yields 5 independent fingerprints.

This inverts the usual relationship between precision and information. More precision in fabrication would reduce security. The optimal manufacturing process is the one that controls nothing.

Note (2026-11-11): checked against the full text of the paper (Xu et al., arXiv 2603.18518, laser-scrawled random plasmonic metasurface physical unclonable functions). The nanosecond laser scribing, the bit uniformity of 0.500, the inter-Hamming distance of 0.499 and the capacity of about 28,000 bits per pattern are as in the paper. Two points differ. (1) The essay says the same disorder 'interrogated at 5 wavelengths yields 5 independent fingerprints'. The paper's five-fold expansion comes from concatenating two keys derived from grayscale (topography) images with three keys from the red, green and blue channels of the pattern; it describes the capacity as achieved 'solely through spectral channel decomposition without multiplexing physical parameters'. That is three colour channels plus two grayscale-derived keys, not five wavelengths. (2) The paper's abstract also reports 'resistance to reverse nanofabrication' and environmental stability, and its security argument rests on the unclonability of the pattern together with measured authentication statistics (a separation threshold between intra- and inter-pattern distances); the essay's statement that 'more precision in fabrication would reduce security' is the essay's inference.